Repository navigation
Termwright v0.36.0
-
Start page: the line beside the name says AI-powered network access, automation, emulation and reservation platform (it wraps in a narrow window instead of being cut off) over SSH · Telnet · Serial · SFTP · RDP · VNC · WoL · Discover · Teams, a little dimmer, and the footer Built for the people who design, operate, and troubleshoot networks. Both are a little smaller and a softer white.
-
About: three more cards - AI Agent (autonomous configuration and troubleshooting in your sessions and labs; MCP connector), Emulate (virtual labs) and Collaborate (team server: shared inventory, who is on a device, reservations, usage) - three to a row in a wider window, a new introduction, raised buttons with icons like the start page's, and Star on GitHub and Share (on LinkedIn) beside Copy Link; Created by moved beside Close, easier to read.
-
Team server: model, serial number and MAC on Inventory. The admin page's Inventory has Model, Serial # and MAC columns (searched and exported to CSV too). The team server reads them itself: it logs in to EXOS, VOSS and Cisco IOS devices over SSH (or Telnet), turns paging off, runs the show commands and logs out, with the Device logins on Discover - one per kind of device, the factory defaults until an owner or co-owner sets others (EXOS admin with no password, VOSS rwa). Devices are read when they are added (from a scan, + Add device or Import CSV; can be turned off), and any time with Read Device Facts for the selected devices. A refused login shows as login failed on the device, with the reason on hover. The logins stay on the team server; members never get them. Members' Termwright also sends the serial number and base MAC (with the hostname and model) when it reads them, so devices are filled in without the logins too. A grey MAC was only seen on the network; the device's own base MAC replaces it once read.
-
Captures (from the user's screenshot): the packet's detail under the table has a ✕ (and Esc) to close it; while a capture runs the table keeps the newest packets in view - at the bottom, followed unless you scroll up to read - and Newest First puts them at the top instead (kept for next time).
-
Open Web Page Through This Session: the page is tried through the session first - when it cannot be reached, Termwright says why instead of opening a browser on nothing (an SSH server with TCP forwarding off, as on EXOS and VOSS switches; a name the device cannot resolve; no answer). A name the device cannot resolve (a switch without a DNS client) is resolved on this computer and reached by its address. Device types have a Tunnel setting: EXOS, VOSS and Cisco IOS do not forward by default, so the menu item is off for their sessions ("no forwarding") and the dialog points to Open Web Interface or a Linux session instead.
-
Emulated lab, Server (from the approved mock-up): a palette node next to the PC - an Alpine computer with HTTP, FTP and TFTP from one shared folder, DHCP on its port 1 (range, gateway, DNS, lease, boot file and TFTP server) and SSH/SFTP; its services answer on the management address too (EXOS:
download url http://<server>/<file> vr VR-Mgmt), DHCP only on the data port. Its side pane: services on/off with their settings, how to reach it, the shared folder (Upload… from this computer, Download, Folder…, Delete with check boxes) and the DHCP leases; service badges under it on the canvas. The first server installs its packages once and becomes the Server image, so later ones start offline. Connector: lab_server (add, show, set, upload, delete). -
Emulated lab, microcontrollers (from the approved mock-up): ESP32-S3, ESP32-C3, ESP32-C6 and ESP32-H2 as lab devices. Add Firmware… (Appliances) takes a flash image or an ESP-IDF build folder; the firmware is drawn as a dev board under Microcontrollers; it runs in Espressif's QEMU on the KVM host (downloaded from its newest release and checked against its SHA-256) with its UART as the console session, its flash kept, a GDB server on request and Flash to another firmware. They now also run on this computer (Hosts → Add → This computer, macOS or Linux: an SSH connection inside Termwright, no port or login; Espressif's QEMU downloaded once), have Ethernet on port 1 for lab cables (Espressif's OpenCores MAC,
openethin ESP-IDF), and the ESP32-C3 offers MicroPython, downloaded from micropython.org when it is first dropped on the canvas (or Add Firmware → Get MicroPython) - the S3 and C6 builds stop in Espressif's QEMU, so they get no sample. Tried on a Mac (macOS 26, arm64): This computer, the QEMU download with its SHA-256, the Homebrew message and the start worked; the console was silent because Espressif's UART drops output while nothing is connected, so Termwright now keeps every board's console from its first instruction (it starts paused) and relays it to the console sessions. The Hosts page shows this computer as a microcontroller host (no KVM checklist or Set Up). The flash size follows the firmware header. The lab's status checks work on a Mac too (sysctl, vm_stat and ps where Linux has /proc). The connector has lab_microcontrollers (list, this_computer, sample, add, flash), and lab_console types on a board's UART without a login. -
Open Web Interface reads the device's web server (the user's request): for EXOS and VOSS sessions the dialog logs in in the background and shows whether HTTP and HTTPS are on; Turn On HTTP and HTTPS… shows the device's commands, runs them and saves the configuration (EXOS:
enable web http, a self-signed certificate,enable web https,save configuration; VOSS:web-server enable,no web-server secure-only,save config), then the address follows (https:// once HTTPS is on). The commands and the patterns that read the state are part of the device type (Global Options → Device Types), so other device types can have them too. Background commands can now answer a device's (y/N) question when the user approved them. -
Emulated lab, from the user's screenshots: in the light theme the switches are dark silver (white on a light canvas before) in every style but Boxes; a CPU meter over 100 % (QEMU's own threads count too) stops at its track and turns amber instead of running over the number; the lab bar shows the switches' states as round count badges (5 up, 1 off…) instead of a bar; a switch's hover card goes where it fits in view - right, left, below or above - never over the switch; the captures list has a check box on each stopped capture and Select all, Delete is red and the confirmation's Delete is red too (no tiny ✕); a running capture's Stop pulses slowly; the management capture badge is blue like a captured cable.
-
Session properties → Jump Host / Proxy: with a jump host or proxy set but turned off, a small square showed on the left of its switch (the greyed section also greyed the switch's hidden check box, making it visible). Fixed.
-
Open Web Interface on a switch that serves only HTTP: the address is offered as https://, but an EXOS switch serves only HTTP (port 80) until HTTPS is turned on (
show management: "Web access: Enabled (tcp port 80)"), so the page would not connect. Termwright now checks the route first: when port 443 does not answer and port 80 does, it opens http:// and says why. -
AI chat finished: a microphone in the message box starts the system's dictation there (macOS Dictation, Windows voice typing with Windows+H; elsewhere a hint); the empty chat explains Agent and Ask; the old layout's leftovers are gone.
-
Session properties: side panel or full window (issue 5; pictures for approval): the welcome screen asks where a session's properties open, with a picture of each - A side panel (the everyday fields, More Options… for the rest) or A full window (every page at once); Settings → General → Session properties changes it later.
-
AI chat, like VS Code's chat (issues 10, 12, 15; screenshots for approval): your messages as bubbles on the right; the agent's steps as one line (Ran 3 commands on core-1, acc-1) that opens to them, its questions always in view; copy, model and time under each answer. The model picker (model · effort, with every version) moved from the top into the message box next to + (add context), the mode, the tools and send; what the agent may do sits under the box. Agent is the default mode and plans: a task of several steps gets a checklist plan, pinned above the message box, that fills in as it goes. Plan mode looks with read-only commands and writes the plan; Run the Plan carries it out. The mode is kept.
-
Open Web Interface reaches the device (issue 6): a device without a jump host or proxy - an emulated switch on the host's network, say - opened in the default browser, which follows the system's or the company's web proxy; that proxy cannot reach a lab or management address, so nothing arrived at the device. It now opens in a browser window of its own whose traffic goes straight from Termwright, the way the SSH session connects (the default browser only when no Chrome, Edge, Chromium, Brave or Firefox is there). Install from ISO: an automatic Alpine install needs a disk of at least 1 GB and makes no swap partition (100 MB and 500 MB disks ran out of space making the boot files, after the erase question was answered).
-
Emulated lab, Traffic → Statistics (from the approved mock): every test port's counters in one table, grouped by host (a virtual chassis each, with its rates and load) - link, frames per second and Mbit/s each way, frames sent and received, loss of the frames that were for the port, latency, errors and the last minute of its rates; a port's streams and a stream's flows below it; Streams and Flows views; live every second, Freeze, Clear counters, Columns, sorting, folding a host and Export CSV. Also in the bottom pane's Traffic tab, in
lab_traffic statsand in Pythonport_stats(). The test ports now count every frame and byte they send and receive, their streams' frames per port, latency and errors. -
Emulated lab, Appliances touch-ups (from the user's screenshots): the bottom pane keeps only its tabs while the Appliances page shows (a click on its tabs or dragging it opens it again; other pages get its height back); the install from ISO form shows its two kinds of install as a switch (they ran together as text), Memory has its MB / GB choice like Disk, Disk, Memory and vCPU boxes are one width, the Check box sits with its text, and labels line up with the top of their fields; the device pack view leaves out the pack's empty entries. Alpine from ISO without internet: the install takes the ISO's own packages first and the release's mirror after them (it failed with "unable to select packages: syslinux" when the installer could not resolve the mirror's name), and the "Erase the above disk(s) and continue? (y/n) [n]" question is answered yes (a default written after the question kept the console waiting).
-
Emulated lab, switches left running (found by the user: the lab icon counted 7 with 3 running): deleting a saved topology now switches off its switches that still run in the background (before, they ran on with no topology to stop them from). Hosts → Running here marks such switches left running and has Stop… for every lab running in the background (asked first); the lab icon's tip says to stop them in Hosts.
-
AI chat: Claude Code sign-in (asked for): when Claude Code is not signed in, the message says what to type on this computer - Open Terminal and run: … on macOS, Open PowerShell and run: & '…' on Windows, Open a terminal and run: … on Linux, with
claudeor its whole path when the terminal would not find it - and offers Run it in a Termwright tab (a Local Shell with the command typed) and Copy the command. -
Emulated lab, Appliances page (asked for, from the approved mock): the Images and Templates dialog is now the Appliances tab next to Traffic - a full page with its views in one switch (Images, Templates, Device pack, On the hosts, Public catalogue), counts at the top (images and families, on this computer, to download, on the host, in use), the images by family with a state badge each (this computer, on the host, to download, downloading, copying, file missing) and quick filters beside the search and sort, and the chosen image in one header (what it is, persona, used by, Download / Copy to Host… / Remove) with its settings in labelled cards: File, Hardware, Console login, First boot, Management address (with the DHCP on commands), Persona and Advanced (name, version, serial format, QEMU options). Templates and installs from ISO use the same layout.
-
Emulated lab, drawing styles (asked for): Classic 3-D is taken out (topologies that used it draw as Rack front) and Rack front (1U) is the default; the hidden Style menu can no longer widen the toolbar (it showed a scroll bar).
-
Emulated lab, Start All and Stop All (asked for): Start All is on only while a switch is stopped, and it starts only the stopped ones - running, booting and failed switches stay as they are (Start never restarts a running switch now, also through the connector and the API; Restart does that). Stop All asks first.
-
Canvas and Traffic touch-ups (from the user's screenshots): on an empty canvas every drawing style can be chosen (Slate tiles, Steel isometric and Rack front sat under the canvas and took no clicks); the device list's search box stays at the top while the list scrolls; the Traffic page's three chart cards have one height, their empty messages are centred, and the test port column lines up with them (its empty note is a dashed card). The device list keeps room for its scroll bar and never scrolls sideways, and the canvas column never grows with what it shows, so a drawing style no longer moves the toolbar. In the side pane only one of a switch's sections is open at a time (Details at first: opening another closes it), and Restart / Configure… sit clear of the line below them.
-
Toolbar shows what is active (asked for): the AI icon turns blue while the AI panel is open or working, the cube while VMs run on any KVM host, the flask while simulated switches run (blue stays under the pointer). In the AI chat your messages show a solid grey person avatar instead of a purple "Me".
-
AI chat: copy buttons (asked for): what you asked, each answer and each command's output have a copy button (it shows when the pointer is over them), as in VS Code.
-
Fixes from the user's list, part 2: Set Up Remote Desktop opened on the last result for good, so a desktop could not be set up again - not the other one (VNC after RDP) and not after its sessions were deleted; the dialog now opens on its form, and a finished setup offers Set Up Again…. The AI agent's Read only never asks: read-only commands run by themselves, production sessions too (nothing that changes anything runs in it). A screenshot on the clipboard pastes into the AI chat with the app's own Paste (⌘V / Ctrl+V, which reached the chat as text only); text selected in an answer is copied with Copy even while the message box has the focus. File Servers' firewall: the command to allow the servers has
sudoon each line and follows the ports as they are typed (it needed Save, close and reopen); Allow Termwright… takes your sudo password on macOS and Linux (empty: the system's own password prompt), and firewalld opens a TFTP port other than 69 by number, now and after a restart. -
Emulated lab, KVM Hosts page touch-ups (asked for): each host shows its state as a square badge (ready, needs set up, N to fix) and its switches as round counts with a white number - green running, orange booting, red failed (a running switch with a problem counts as failed; switches of other topologies count as running); the CPU and RAM meters sit on one line; the set-up output has Copy Log and a real Close button.
-
Emulated lab, switches that failed on the host's network (found in a lab of seven switches): a switch whose management port never asked DHCP waited 2 minutes and failed with "no DHCP address" - an image can now name DHCP on commands (
dhcpOn, brought by the device pack), typed when the switch has no address yet; the failure message no longer blames only a VMware host's virtual switch. A pager line that reached Termwright in two parts ("Press ", then the rest) was taken for the switch's prompt - it ends in ">" - so the address command's output was cut short and the next command typed into the pager; it now waits for the whole line and pages on. A switch found running when Termwright starts, on the host's network with its address not known yet, has its address read again (it stayed "DHCP (not known yet)"). A DHCP on step that fails is shown in Events. A login that is not ready yet ((pending-AAA) login:) and then only says it is ready, without a new prompt, gets another Enter instead of failing with "no login prompt" after a minute (switches booting together on a busy host). -
Emulated lab, open items closed: a rectangle drawn on the canvas now also selects test ports, clouds, hubs and L2 switches (N items selected), and Delete removes them all; the Log tab is a table (time, source, message) and Captures uses the same search box; the lab page loads after the rest of the app, so the app starts faster; an Alpine installed from ISO no longer gets the installer's SSH server and NTP client when the answer file says none (setup-alpine ignored a bare "none" and installed openssh anyway); on Windows
termwright lab …prints its output in the terminal it was started from; the connector lists the running app's own tools (the newest ones,lab_install_*included) instead of the list built into an older copy. -
Emulated lab, fixes from the user's list: each topology's sessions go in its own folder under Emulated Lab, with Mgmt (SSH/telnet to the management addresses) and Console (serial consoles) under it; New Topology puts hosts (Linux computers) on the edge switches - in leaf-spine on the leaves, one each in turn, drawn in a row below - and shows them as Host; New Topology has a working Device icons choice, and the toolbar's own menu now says Buttons with text / as icons only / with labels below (it changes the toolbar, not the devices); a capture on a management port shows a red ● mgmt badge on the device (click for the packets); the cloud is silver instead of yellow.
-
Emulated lab, a Linux PC with a web browser (asked for): a Linux computer can get a desktop - X, a small window manager and Firefox (maximized, opening the page you give, for example a switch's web interface on the data network), started at every boot - used through its Screen: clicks go to its mouse (a USB tablet, added at its next restart), typing to its keyboard. It needs internet to install and 1 GB of memory. Connector
lab_computeractiondesktop,lab_keyswithclick; Pythondesktop()andclick(). -
Emulated lab, route generator: OSPF and BGP with many routes (asked for, protocols part 2): a Linux computer of the lab installs FRR and speaks OSPF (an area on one of its data ports) or BGP (a neighbour, its AS and the switch's) with the switch, advertising up to 10,000 prefixes from a first one (100.64.0.0/24, 100.64.1.0/24, …). It answers ping on every address in them, so an ICMP stream from a test port to the prefixes - one address in each, in turn - shows whether the switch forwards to every route it learned (loss per run, as with any stream). The neighbour's state (Full, Established, …) and the prefixes are read back from FRR. Connector
lab_routes(set, show, off, stream), Pythonlab.routes(...); the routes come back when the computer restarts. -
Emulated lab, built-in nodes: PC, L2 switch, hub and cloud (asked for, built to the approved mock-up): Built in · always here in the palette has the four, each with its own icon. PC adds a Linux computer (Alpine first, iperf3 installed). L2 switch: 8 ports to start, 8-24 with +/− in its panel; VLANs per port (access, or trunk with tagged VLANs and an optional native one) set in its panel or on its command line; an in-band management address that answers ARP and ping; a console session (telnet through its KVM host) with a command line (
show ports,show vlan,show fdb,configure port 1-2 trunk vlans 10,20,configure management ipaddress …), and SSH when the host's Python has paramiko; its panel shows the ports with what is on them, their VLANs and links, and the MAC table. Hub: 8 ports, no management. Cloud: one per host with 1-8 ports (1 to start, +/− in its panel), so the canvas does not fill with clouds; a cable to a full cloud is refused. Connectorlab_hub(show, ports, port, management, cloud_ports) and Pythonhub()do the same. -
Emulated lab, KVM Hosts redesigned (asked for, built to the approved mock-up): the host list on the left scrolls while + Add and a red − Remove stay fixed at the bottom left, with Images (All Hosts) and Close in the same bottom bar. Add has no device-type field any more: it shows the Linux SSH sessions as the Session Manager has them, by folder, searchable by name or IP address, several at a time. The host's details have Images at the top (where Remove Host was): a page of that host's images with type, version, size, the devices using them and copy progress, Copy to… another host and Remove from this host (only when nothing there uses it). Images (All Hosts) is the hosts × images matrix with Copy here per cell and Copy Missing to All…. New connector tool
lab_host_images(matrix, copy, remove) and Pythonhost_images(). -
Emulated lab, clients behind a test port: 802.1X, DHCP and IGMP/MLD at scale (asked for: "the traffic generator itself can simulate dot1x clients (n number of clients)"). A test port can stand for up to 4,000 hosts, each with its own MAC (counting up from 02:63:…, or a first MAC you give), started a few a second (50 by default). Each logs in by 802.1X (EAPOL-Start, then EAP-MD5 with a user name and password;
{n}is the client's number and{mac}its MAC, souser{n}makes user1, user2 …), takes an address by DHCP once it is in, answers ARP for it, and joins multicast groups (IGMPv2 / MLDv1, again on every query). The counts - with an address, logged in, refused - and the first 256 clients with their address and 802.1X state come back with the traffic status. Set them with the connector (lab_trafficadd_port / set_portclients), the API (clientson a test port) and Python (add_port(..., clients=dict(n=200, dhcp=True, dot1x=dict(user="user{n}", password=…))),traffic.clients(name)). The screens for it follow the mock-up. -
Emulated lab, a built-in RADIUS server (planned with the 802.1X clients): it runs on every host of the lab and answers the switches there - EAP-MD5 (with State and Message-Authenticator), PAP and MAC authentication (MAB, by MAC patterns such as
02:63:*); accepted users get their VLAN (Tunnel-Private-Group-Id). Users can be patterns (user*). It also takes accounting on the next port and keeps the last 200 logins. Connectorlab_radius(show, log, set, add_user, remove_user; secrets and passwords only withreveal), API/lab/tools/radius, Pythonlab.radius(...). It listens on the lab bridge's address (bridge mode) or the host's. -
Emulated lab, spanning tree on test ports (STP / RSTP) (asked for, protocols part 3): a test port can send configuration or RST BPDUs every 2 s as a bridge with a priority you choose (lower than the switch's makes it the root), signal a topology change on request (
topology_change), and reads the switch's BPDUs back: the root the switch announces, its own bridge, its path cost, the port role (RSTP) and whether it takes the test port as the root. -
Emulated lab, QoS under congestion in the test library (asked for): A and B send into C at 130 % of line rate together - high (EF, PCP 5) 10 %, medium (AF41, PCP 4) 30 %, best effort 90 % - and it passes when the high class loses nothing and keeps its latency (≤ 5 ms) and the medium class loses ≤ 0.5 %; best effort takes the loss.
-
Emulated lab, KVM Hosts → Add a Host lists only Linux machines (asked for): only SSH sessions whose device type is Linux (or a Linux distribution's name) are offered; without one it says to make one in the Session Manager. Remove Host… is red. Buttons marked as dangerous but outlined (Cancel an install, Remove a hub) are now red too - their style was missing.
-
Emulated lab, protocol emulation, part 1: LACP and LLDP on test ports (asked for; the order agreed: LACP + LLDP, then OSPF and BGP, STP/RSTP, IGMP/MLD and DHCP at scale). Test ports with the same LAG name speak LACP (802.1AX, active; fast = every second, else every 30 s) and form one link aggregation with the switch: they answer its LACPDUs, go in sync, then collecting and distributing; their cards and the editor say LAG lag1 · up and the switch's port and key. A stream to one member of a LAG is received on any of them (the switch hashes). Test ports now also read the switch's LLDP: its name and port show on the card (⇆) and in
lab_traffic list. API/Pythonlag,lacpFaston test ports; connectorlag,lacp_fast. -
Emulated lab, hub and L2 switch in the palette (asked for): Built in in the device palette has Hub (what comes in on a port goes out of every other one) and L2 switch (a plain switch without VLANs that learns where each MAC address is; entries age out after 300 s) - lab nodes without a VM, up to 16 ports each. Drag one onto the canvas or double-click it; its panel adds switch ports to it (any free port), takes them off, renames it or removes it. Its ports work at once, also on running switches:
hub.pyon the host joins them; a hub's switches run on one host. Connectorlab_hub(list, add, connect, disconnect, remove), Pythonlab.hub(...). -
Emulated lab, Overview (asked for): a page next to Canvas with every device in one table - state, kind and version, host, management address, CPU and memory now, cables, checkpoints and what needs attention (a problem, a restart needed) - with search, filters and sorting, a summary (running, stopped, restart needed, with a problem), and per row Console, Screen, Restart, Stop or Start; a row shows its device on the canvas.
-
Emulated lab, the command palette (asked for): Ctrl+K (⌘K on a Mac) opens it - type a few letters of a device, a page or an action (Start All, Add a Note, Add an Area, New Stream, Test library, Changes, Events; per device: show it, Console, Screen, Checkpoint Now, Restart or Start) and Enter runs the first; ↑ ↓ choose, Esc closes.
-
Emulated lab, notes and areas on the canvas (asked for): Note and Area in the canvas bar. A note is text on the canvas (double-click to write); an area is a named, dashed rectangle behind the devices - a site, a rack, a test bed. Drag them by their title, resize from the corner, pick one of the eight palette colours, ✕ or Delete removes the selected one. They are kept with the topology. Cables no longer catch clicks between them (the links layer lets clicks through to what is under it).
-
Emulated lab, a device's screen and keyboard (asked for: "VNC console"): Screen… in a device's menu on the canvas and on a running install shows what the VM's display shows (read through its QEMU monitor, refreshed every 1.5 s) with a line to type on its keyboard (Enter sends it with Enter) and keys - Enter, Esc, Tab, arrows, F2, F12, Ctrl+Alt+Del - for installers and systems that use the screen rather than a serial console (no VNC port is opened). The API has
GET /api/v1/lab/screen/{device}(orinstall:{id}) as PNG; the connector'slab_keystypes on a device's or installer's keyboard; Pythonlab.screen(),lab.keys(). -
Emulated lab, typing a line into a running install (asked for): the screen's Serial console box, the connector's
lab_keyswithinstallandconsole, and Pythonlab.install_type(id, line)send a line (and Enter) to an installer's serial console - while it installs, or while a manual install waits. -
Emulated lab, a verification boot after Install from ISO (asked for): when the image is made, Termwright boots it once on a throwaway layer (the image is not changed) and waits up to 5 minutes for its login prompt; the install then says "Verified: it booted to its login prompt in 38 s", or why it is not verified (the image is kept either way). On by default; Check in the install form turns it off (API
noVerify). -
Emulated lab, checkpoints: typed commands can be undone (asked for). Before Claude or the API types on a running lab switch (
send_commandin a tab of it,lab_console), Termwright takes a checkpoint - the switch's whole running state kept in its disk (a QEMU snapshot taken while it runs) - and the change in the history says so; Revert restores it, so the switch is back to just before (configuration, tables, sessions). At most one automatic checkpoint per switch every 2 minutes, none while traffic runs (it pauses the switch for a moment), the last three kept; a switch in Changes turns them off. A switch's panel lists its Checkpoints with Restore and ✕, and + Checkpoint Now. Connectorlab_checkpoint(list, take, restore, remove, auto); Pythoncheckpoint(),checkpoints(),restore(),auto_checkpoints(). -
Emulated lab, hot-plugged cables (asked for): a cable added between two running switches on one host works at once - the host's patch panel (
patch.py) joins the two ports until the switches restart with the cable; a cable removed from running switches is pulled at once (both links down). Between two hosts a new cable still needs both switches restarted. -
Emulated lab, a test library for the traffic generator (asked for): Library… on the Traffic page opens ready tests - unicast both ways, IMIX both ways, voice/video/data QoS classes, MAC table fill (4,096 MACs as flows), round-trip latency, broadcast containment, RFC 2544 throughput and latency, RFC 2544 back-to-back and RFC 2889 address learning. Each names its test ports by role (A, B, C); choose a test port per role and Apply adds its streams and tests (names already taken get a number). Save as Template keeps the lab's streams and saved tests as a template of yours (their test ports become the roles), for any topology. The connector's
lab_traffichastemplates,apply_template,save_templateandremove_template; Pythonlab.traffic.templates(),apply_template(),save_template(). -
Emulated lab, pcap replay (asked for): a stream of type PCAP replays the frames of a capture file (classic pcap; pcapng is refused with how to save it as pcap), at a rate or with the capture's own timing, through it once or several times (Loops), as captured or from the test port's MAC to the destination's. The capture is kept by Termwright (by its contents) and copied to the hosts with test ports. The receiving test ports know its frames by their contents, VLAN tags left out, so they count them (and the VLAN/priority/DSCP they arrived with) on switched paths. API/Python fields
pcap,pcapTiming,loops,rewriteMacs; connectorkind: pcapwithpcap,pcap_timing,loops,rewrite_macs. Test ports on several hosts already worked (streams between hosts run over the cables between them); round trip (above) gives their latency. -
Emulated lab, a "restart needed" badge (asked for): a running switch or computer whose vCPUs, memory or disk were changed in Configure… shows ↻ next to its name on the canvas, restart needed in its panel title, and a box under its buttons that says what a restart changes ("vCPU 2 → 4 · RAM 2 GB → 4 GB"); it runs with its old settings until then. CPU requests and limits still apply at once, without a restart.
-
Emulated lab, the lab's name servers (asked for): Name servers in the Hosts popover (the toolbar's hosts button) sets up to three DNS servers for the lab (empty: 8.8.8.8 and 1.1.1.1). Computers that have none get them when they start; images' day 0 can use
{dns},{dns2}and{dnsList}; Alpine ISO installs use the first. Linux images whose day 0 had the fixed servers take the lab's from now on. The connector haslab_dns, Pythonlab.dns(...); a change is in the lab's history and is reverted with it. -
Emulated lab,
lab_imagesfilters by name and device type (asked for):name(part of the id, name or version) andtype(EXOS, VOSS, Linux…) keep the answer short; Pythonlab.images(name=…, type=…). -
Emulated lab, narrower vCPU and Memory boxes in Install from ISO and the image settings (asked for), and the Traffic page has its own component with its own styles (
traffic-page.ts; the lab workspace's styles shrink by a fifth of what the page used). -
Emulated lab, traffic generator measurements on par with hardware testers (asked for: the Ixia gaps). Rates in % of line rate or Mbit/s: the stream editor's Rate has a unit (pps, Mbit/s, % of line); Mbit/s and % count the frame plus its 20 bytes of preamble and gap, as testers do, and the editor shows the frames per second it comes to (a test port sends at most 100,000). A test port has a Line rate (Mbit/s, default 1000); streams in % follow it when it changes. Latency percentiles and histogram: every frame's latency goes into a quarter-octave histogram, so each stream has p50, p95 and p99 (within 10 %) besides min, average, max and jitter; choosing a stream on the Traffic page shows them with the histogram. Per-flow counters: a stream with changing fields (ports, addresses, MACs, VLAN) counts each combination of their values as a flow, up to 1,024 flows: sent, received, loss and latency per flow, the most lossy first. Tracked VLAN, priority and DSCP: the receiving test port counts the VLAN, 802.1p priority and DSCP each frame arrived with; frames remarked on the way stand out. Round trip: a stream can have the test port it goes to send each frame back, so its latency is the round trip measured on one clock - which also gives latency between test ports on two hosts (streams between hosts already ran over the cables between them; their one-way latency cannot be measured). Report graphs: the HTML report has throughput, loss and latency over the run, and per stream its latency histogram, arrival marks and flows; the CSV and text reports have p50/p95/p99, one way or round trip, and the flow counts. The API and Python take
rateUnit/rate/rtton streams andspeedon test ports; the connector'slab_traffictakesrate,rate_unit,round_tripandspeed. -
Emulated lab, the two-computer test passes from Python on the real app (
example_two_computers.py: ping both ways 0 % loss, iperf3 28.6 / 28.8 Mbit/s, ALL PASS). Bug fix found there:lab_computer set(and Pythoncomputer_set) answered with the address the computer had before - it is typed on the console and read back a few seconds later - so pc-1 showed "no address" although it had 10.20.0.11. It now waits for the reading (up to 25 s) and answers with what the computer has; if the reading does not come, it says so. A stopped computer's settings are saved and typed when it starts. -
Emulated lab, bug fix (found running the two-computer test from Python):
lab_waitand Python'swait()never saw a running device as ready and always waited until their timeout (10 minutes by default): they wanted an empty state text, but a running device's text is "running". Ready now means running with nothing left to do (not booting, typing its day 0 or moving) and no problem. -
Emulated lab, the two-computer test ran on teams-vm through the connector (asked for): pc-1 and pc-2 (Alpine) cabled to 5520-24t-1 ports 10 and 11 (Default VLAN; the switch's 10.10.0.0/24 is its MLAG ISC VLAN, so the computers got 10.20.0.11/24 and .12/24), ping both ways with no loss (1.5-7 ms), iperf3 19.7 Mbit/s pc-1 → pc-2 and 23.7 Mbit/s pc-2 → pc-1 through the switch's software data plane, the computers' MACs learnt on ports 10 and 11.
scripts/python/example_two_computers.pyruns the same test from Python. Bug fixes found on the way: a device's last hour of CPU (panel, CPU tab, lab_cpu) was never shown because it was filled in before the device's running state was known. -
Emulated lab, bug fix (found running the two-computer test): a Linux computer from the public appliance list had no data ports to cable - the registry's Alpine image has one network card (its management), so pc-1 and pc-2 could not be cabled to a switch. A Linux image with fewer than three network cards now gets the computer settings (three cards: the management port and two data ports) when Termwright starts; a running computer gets its new ports when it restarts. The Events say a CPU limit held a device back only when it matters (ten times or a second in ten minutes), not for every short hold.
-
Emulated lab, bug fix (found on teams-vm): pc-1 and pc-2 could not reach the Internet by name - their routes and the Internet worked (ping 8.8.8.8), but /etc/resolv.conf was empty: the Alpine answer file gave setup-dns its server without
-n, and an installed Alpine with a static address gets no name server from DHCP. The answer file now saysDNSOPTS="-d lab -n 8.8.8.8", and a Linux computer's day 0 writesnameserver 8.8.8.8and1.1.1.1when it has none - before it installs iperf3 (Linux images made earlier get the line when Termwright starts). The connector's tool descriptions no longer say that lab changes are asked. -
AI panel, the in-app AI agent no longer asks before lab changes either (asked for: "stop asking in this AI panel too for lab devices"): its lab tool calls and its own lab commands (start, stop, cable, pull, LAG…) run at once and are kept in the lab's Changes history as AI agent, with Revert; Read only still refuses them. Its commands to real sessions keep their own permission levels, and production sessions are still asked.
-
Claude connector, lab changes are no longer asked; they are kept in a change history with Revert (asked for: "do not ask permission for the lab networks … stored in the history … so that we can revert … only for the real sessions and marked as production we should be very careful", for automated testing and debugging). Every lab tool change by Claude (the connector, the AI agent) or the API is kept with the lab as it was before (lab-history.json, the last 200), and so is each command Claude types in a tab of a lab device. A new Changes tab in the bottom pane lists them (time, by, tool, what, result; search and sort) with Revert to before: it first shows what it will do, then removes the devices added since, puts back cables, CPU requests and limits, computers' addresses, test ports, streams and positions, and starts or stops devices as they were; commands typed on consoles and devices removed since are said, not undone. The connector has
lab_history(list, plan, revert) and Pythonhistory(),revert_plan(id),revert(id). Tabs of real sessions are still asked; sessions marked production are asked for every command and every connection, with a PRODUCTION title, never "for 15 minutes". Never in Settings still stops everything. -
Emulated lab, bug fix (found on teams-vm): a Linux computer whose image type is written "LINUX" stayed booting although its console showed a shell prompt: the check for a computer that logs in by itself compared the type case-sensitively. It now takes Linux in any case, and any image with port address commands (a computer).
-
Emulated lab, polish asked for after the mock-ups: the canvas toolbar's toggles that are on (Grid, Ports, Addresses) show only a coloured icon and label, no filled box; the device panel's state (● RUNNING) is smaller and sits at the right, so the device name has the room (a long name ends with …); the bottom pane's stream sparklines use the Traffic page's colours (sent blue, received orange). Bug fixes: a test port could not be changed once made - a click on it on the canvas (or on its card on the Traffic page) now opens its settings (name, address or DHCP, gateway, VLAN, IPv6 or SLAAC, LLDP, multicast) with Save; the Add button of a new test port, and Save / Save and Run of a stream, were at the bottom of a scrolling editor - they are now in the editor's title row. A test port is shown as "core-1 · port 5".
-
Emulated lab, a Traffic page next to Canvas (asked for: "traffic next to canvas as a separate page"): the traffic generator was only a tab of the bottom pane. The page has the buttons at the top, seven tiles (test ports and their switches, streams and how many run, sent and received frames per second now, loss so far with the frames lost, average and highest latency, streams passed and failed - a warning or a tick, never colour alone), the test ports as cards, three live charts of the run (throughput sent/received in pps, loss so far per stream, latency per stream; one axis each, a legend, 2 px lines in a colour-blind-safe palette checked for the dark and light themes, a crosshair with the values), the streams table and the stream and test-port editors, and a Tests view with the RFC 2544 tests. The traffic history now keeps each stream's loss and latency with its rates (REST/connector
history: [seconds, tx/s, rx/s, loss %, latency ms]). -
Emulated lab, CPU request and limit as kubectl counts them (asked for: "more clear and clean … like docker or kubectl, 20m cpu"): the device panel's CPU share (Low/Normal/High) and CPU limit (cores) are replaced by a request and a limit in millicores (
250m, or0.25cores; empty = none). The request is its share when the host is busy and reserves nothing - it may use more when the host is free; the limit is the most it may use. A bar shows its use now against its vCPUs with the request and the limit marked, the last hour (average, p95, peak), how much of the time and how many times it went above its request, and how many times and for how long its limit held it back (read from the cgroup's cpu.stat on the host); a device often held back gets advice and a button that raises the limit or the request, and an Event at most every ten minutes. The panel now says that the request and the limit apply at once with no restart, and that vCPU and RAM (Configure…) are used from the next start. A new CPU tab in the bottom pane reports every device (sortable, filterable) with the totals per host - requests against its cores, limits, use now. Connector/RESTlab_cputakesrequest_mandlimit_mand reports the last hour; Pythoncpu(request_m=, limit_m=); earliermax_cores/sharesettings carry over (share 50 = 500m). -
Lab windows, search, filter and sort in the remaining lists (asked for): the image list, the template list and the host list have a search box at the top (it stays there while the list scrolls; "N of M" when it narrows the list) and a row of sort and filter chips under it (images: Name, Type, State - downloaded, on this computer, on the host, missing - and Size; templates: Name, Image, Model, vCPU, RAM; hosts: Name, State, OS, Machine). A chip sorts by its field, again the other way; its ⏷ ticks the values to show; active filters show as chips with ✕. Images stay grouped by type, sorted inside each group, and the groups keep their place. The switch list of the Lab window (simulated and emulated) has the same search bar and sortable, filterable column titles as Events and Inventory, and Select all ticks only the switches shown. A list that a search empties says so instead of "No images yet". The search box of every table now uses the width it has (up to 320 px).
-
Emulated lab, the device panel has only Restart and Configure… (asked for): the Console and SSH buttons are gone from the right-hand panel; a device's console and SSH open from its right-click menu or a double-click on the canvas, as before.
-
Images and Templates, Install from ISO shows its progress (asked for): an install had only one line and a console tail that ran off to the right as a single line, so a long install looked stuck. Each install card now has a bar and the steps of the install (copy the ISO, start the installer, boot, log in, install, power off, make the qcow2, copy it to this computer; six steps for an install by hand) with the time each step took and a clock that ticks, the size written to the new disk and the conversion's percent (read from the host every few seconds), and the last lines of the installer's console as readable lines; a console that has shown nothing for 90 seconds shows a warning. The list entry says step 5 of 8. The console text no longer keeps pieces such as
[97mwhen a colour code arrives in two parts, and Alpine's own question Setup a user? is answered no (the answer file also saysUSEROPTS="none"), so the install cannot wait there. The disk size can be given in MB or GB (the box is narrower, with a unit list; the API hasmband the connectordisk_mb; at least 100 MB), and a hint says the size is a maximum: the qcow2 file grows only as the system is written.lab_install_statusand the REST status tell the step, the size written and how long the console has been quiet. -
Emulated lab, a Linux computer that logs in by itself stays "booting" no more (found on teams-vm): an Alpine installed from an ISO gives root a shell on the serial console at once, with no
login:line, so the computer stayed booting for ever although its console worked. A Linux computer is now booted also when a shell prompt waits at the end of its console, and an Alpine installed by Termwright gets the ready textlogin:|Welcome to Alpine. -
Images and Templates, Install from ISO (asked for): some distributions (Alpine Linux, for example) are published only as an installer ISO, and an ISO is not a bootable disk image. Install from ISO… (at the bottom of the image list) copies the ISO to a KVM host and runs its installer there, in a QEMU of its own with a blank disk (8 GB by default, with memory, vCPUs and host to choose), then keeps the installed disk as a new image: a compressed qcow2 file kept on the host and on this computer, with the settings of a Linux computer (login root, serial console, port addresses), so it can be a computer on the canvas at once. Alpine Linux installs by itself: Termwright logs in on the installer's serial console, writes the answer file, runs
setup-alpinewith it (the blank disk is erased and installed from the ISO's own packages), sets the root password you give and powers the machine off; the end of the installer's console shows while it runs. Other ISOs: the installer starts and its console opens as a session; you install by hand (the installer must talk on a serial console; at its boot menu addconsole=ttyS0if it does not), then Finish and save as image. Installs are listed with their progress, can be cancelled (the disk is thrown away, the ISO stays on the host) and removed from the list; one install runs per host at a time. Also cloud images with a seed made by Termwright: Public appliances has Debian 13 and 12, Ubuntu 24.04 and 22.04, Rocky Linux 9 and AlmaLinux 9 as the distributions' own qcow2 cloud images, with a cloud-init CD-ROM that Termwright writes itself (a small ISO labelledcidata; nogenisoimageis needed on any host): login root / termwright, serial console. Through the API and the connector:lab_install_iso,lab_install_statusandlab_install_finish(REST/lab/tools/install_iso,install_status,install_finish; OpenAPI updated), andLab.install_iso(),install_status(),install_finish()andwait_install()in the Python client. Russian strings added. The Alpine typing is tested against a simulated Alpine console and the whole job against a real SSH server and QEMU in software emulation; it has not yet run against a real Alpine ISO. -
Emulated lab, a cleaner device panel (asked for): the selected device's panel has its state at the top with Console, SSH, Restart and Configure… buttons, then four folding sections - Details (open), Resources, Runs on (with Move to another host… as a link) and Cables - and a computer's Network; one value per line. Which sections are open is remembered while Termwright is open.
-
Emulated lab, the toolbar above the canvas sat 8 px too high and 8 px too wide on each side in the glass style, so it overlapped the title row, the left pane's scroll bar and the right pane. It now sits between the panes, and all its buttons have the same size (Grid, Ports and the others).
-
Emulated lab, search, filter and sort on tables (asked for): a search box above the Events, Inventory and Traffic tables (any column, with "N of M"), click a column title to sort (ascending, descending, off; Shift-click adds a second key; numbers sort as numbers), and the ⏷ beside a title to show only some values (Level, Switch, Host, Trigger, State, Result) or rows that contain some text; the active filters show as chips with one Clear filters. In Events the Level is a small coloured pill and the Message is in the normal font, coloured only for failures.
-
Emulated and simulated lab, select several devices on the canvas (asked for): drag a rectangle on the empty canvas (or Shift/⌘/Ctrl-click, or ⌘/Ctrl+A) to select devices, drag one to move them all, and press Delete to remove them after one question that names them. Delete now works for one selected device too (it only worked for a Cloud).
-
Images and Templates, the buttons stay at the bottom: Add Image…, GNS3 Appliance… and Import Profiles… no longer sit at the end of the image list (so you had to scroll down to reach them); the list scrolls above them.
-
Emulated lab, Tiny Core is not a computer (first real test: two Tiny Core machines ran but their console stayed blank, so Termwright could not log in or type an address): Tiny Core boots to a screen only and has no serial console. Add Computer now skips it and takes Alpine (or another image with a serial console); a GNS3 Tiny Core appliance is no longer set up as a Linux host.
-
Emulated lab, names with a comma (found on the first real test of two computers):
tc-1,tc-2was taken as one name, and QEMU read the comma in-nameas an option separator, so both machines failed to start ("Invalid parameter"). A comma-separated list of names now names the machines one by one (Claude'slab_add_switchesand the New Lab dialog), and a comma in a name is escaped for QEMU. -
Python client for the lab and the traffic generator (
scripts/python/termwright_lab.py, standard library only, with a guide indocs/PYTHON.mdand an example): the lab tools, test ports, streams (speed in Mbps or percent of line rate, or packets per second),run_streamswithassert_pass, saved tests (rfc2544,convergence,iperf), and reports in JSON, CSV, HTML or text; it finds the address and key itself. Tested against a fake API and against every path and field in the OpenAPI document. -
Emulated lab, Linux computers found by a first test on a real host: GNS3's Linux images (Alpine, Tiny Core…) have one network card, so a computer had no data port to cable; they now get three (management and two data ports, named eth1, eth2) - images added before are corrected when Termwright starts. A console that logs in as a plain user (Tiny Core's tc) is made root right after the login (the image's elevate command,
sudo -s), and Tiny Core gets iperf3 from its own package loader. Add Computer now prefers a Linux image whose file is here over one that still has to be downloaded. -
Settings → AI Assistant → Claude connector: Claude Code showed Not added every time Settings was opened, even after Add worked, because the page never asked. It now reads Claude Code's own settings file (
~/.claude.json, orCLAUDE_CONFIG_DIR) and shows Added when Termwright is in it. -
Emulated lab, a switch that was running when Termwright started (VOSS 5520 stuck at its login with "planned, not on the switch yet"): nothing watched its start, so its day 0 and its fixed management address were never typed after a restart of Termwright, and Type Day 0 Again only cleared its flags. Now a running switch whose day 0 or fixed address was never typed is found at the first look and typed on its console (Events: "found running without its configuration typed"), and Type Day 0 Again really types it. lab_debug says so (a DIAGNOSIS line), names the address typed to it, and ends with the emulator log lines of that switch.
-
Emulated lab, lab_debug: one switch's state, the step Termwright is on, its error and addresses, its image's login and ready settings (never the password), its Events and the end of its console, in one answer for the Claude connector, the API and
termwright lab debug switch=<name> lines=300. Passwords and keys are hidden. -
Emulated lab, a computer works like a PC on a desk: a double-click (or Open Desktop in its details) opens its window with Desktop, Config and Ports. The Desktop has IP Configuration (per port: IPv4 DHCP, Static with address and subnet mask, or Off; default gateway and DNS server; IPv6 Automatic from router advertisements, Static or Off; the link-local address; Apply types it on its console and shows what the computer then has), Ping (IPv4 or IPv6), Traceroute, iperf3 (be the server, or test to one), Web Browser (the page's text as the computer gets it, or its source), DNS Lookup, Capture (a port's packets) and Command Prompt (its console). Computers are drawn as a screen with a tower in every icon style, labelled "PC ·
" with their address/bits. Connector, API and command line: new lab_computer tool (set, or run an app and read its output).
-
Emulated lab, details pane and right-click menu: the details show where a switch runs (Runs on: host, session, address, management network and the host's load) with Move to another host; Configure… is its only button. Console, SSH and Save Console Log moved out of the pane (the right-click menu has Console and SSH, the Console Log tab has Save), and the right-click menu no longer has Configure, Save Console Log or Move to Host.
-
Emulated lab, CPU limits and shares: each switch or computer can be held to at most some host cores (0.5 = half of one) and given a share of the host's CPU when they compete (100 by default, 50 half as much, 400 four times), through a cgroup of its own on its KVM host (cgroup v2 with sudo; without them a lower share still works as a lower priority). A running one takes them at once, a stopped one when it starts; the Events say how. In a switch's details: CPU share (Low 50, Normal 100, High 400) and CPU limit under Resources. Connector, API and command line: new lab_cpu tool (
termwright lab cpu switches=leaf-1 max_cores=1 share=50; without values it lists them). -
Emulated lab, Computer: the canvas toolbar has Add Computer (pc-1, pc-2… from your Linux image, Alpine first), computers have their own drawing in every icon style (a screen with a prompt) and show their data address under the name, and their details have a Data ports section: Fixed, DHCP or None per port, with a ✓ once the address is read back on its console (a DHCP lease shows there). Draw As has Computer too.
-
Emulated lab, Linux computers are configured like switches: an Alpine, Debian, Ubuntu, Fedora, AlmaLinux, Kali or Tiny Core image (from the catalog or a GNS3 appliance; images added earlier too) is logged in to on its console, named on day 0, gets iperf3 when its package manager can fetch it, takes a fixed management address when the lab uses them, and gets the addresses you give its data ports (fixed IPv4/IPv6 or DHCP), typed at each start and at once while it runs. The port is found by its MAC address, so it works whatever the image calls it (eth1, ens4…). An iperf3 test without a server address now uses the server's first fixed data-port address. Connector, API and command line: new lab_address tool (
termwright lab address action=set switch=pc-1 port=1 address=10.10.0.11/24). -
AI chat, the whole emulated lab for the agent: with the Emulated lab tool the in-app agent now has every lab tool of the Claude connector - add switches from images or templates, cables, power, wait until ready, type on a switch's console, check the cabling, move a switch, traffic test ports and streams with their pass/fail report, packet capture, the link check, the QEMU command line and the Events. Reads run at once; each change is asked in the chat with what it does, with Allow for This Chat to stop asking for the rest of that chat (not for changes that may interrupt traffic). Connector: new lab_link_check and lab_qemu_command.
-
Emulated lab, Captures: the Stop button of a running capture is red. A cable being captured keeps its colour (it turned red and looked down): blue dashes run along it and a blue dot pulses at the captured port; captures started earlier show on the canvas without opening the Captures tab.
-
Emulated lab, traffic generator, step 1: streams get frame sizes that step, vary at random or follow IMIX (64/594/1518, 7:4:1), DSCP and 802.1p marking, bursts, and fields that change from frame to frame (source/destination MAC and address, VLAN, ports: stepping or random - hundreds of flows from one stream). IPv6 streams between test ports (each answers neighbour solicitations and ping at its IPv6 address) and multicast streams to the test ports that joined the group (they send IGMP/MLD reports, so snooping switches forward). Each stream now counts out-of-order and duplicate frames, latency min/avg/max and jitter (pass/fail on jitter too), and the Traffic tab graphs its sent and received rate while it runs; the report has the new columns.
-
Emulated lab, traffic step 3: a test port can take its address by DHCP (
dhcpas its address: the lease and gateway show on it and its streams use them) and its IPv6 address by SLAAC (slaac, from router advertisements; it answers neighbour solicitations for its link-local address too), sends a gratuitous ARP when it has an address, and announces itself by LLDP every 30 s (on for new test ports: the switch sees Termwright Test port 1, and Check Cabling can see it). A new test, iperf3 between hosts, runs iperf3 on two host VMs of the lab (Alpine, Debian…) through their consoles - TCP or UDP, target rate, parallel streams, reverse - and reports the rate, retransmits, jitter and loss. -
Emulated lab, traffic tests (step 2): the Traffic tab's Tests runs saved tests as a series of trials - RFC 2544 throughput (the highest rate without loss per frame size, by halving), latency (min/avg/max and jitter at that rate), frame loss (100 %, 90 % … of the top rate until two steps lose nothing) and back-to-back (the longest burst without loss), RFC 2889 MAC learning (how many source MACs are learned before frames flood to a third test port), and convergence (traffic while a cable or LAG member is pulled, or a switch stops or restarts, plugged in or started again after a hold: frames lost and the outage in ms, each way). Results fill in as each step ends, every trial is in its log, and the report saves as HTML or CSV. The same through the connector (lab_test), the in-app agent, the local API (
/api/v1/lab/traffic/tests,…/runs) andtermwright lab test …. A refused console login now says how to give Termwright a password changed on the switch. -
Local API: described by an OpenAPI 3.2.1 document (
docs/api/openapi.yaml, generated from the code and checked by the tests; also served at/api/v1/openapi.jsonand.yaml), resources with the usual methods and status codes (201 created, 204 deleted, 405 with Allow, 422 when the lab cannot do it), RFC 9457 problem details for errors, unknown fields refused; tools at/api/v1/lab/tools/{tool}; a guide indocs/API.md. A test port's IPv6 address and multicast groups are kept when it is changed. -
Local API and command line (Settings → AI Assistant → Local API): the emulated lab for scripts, test frameworks and CI on this computer - a REST API (
/api/v1/lab…: the lab, every lab tool, test ports, streams, start, stop and the report as JSON, CSV, HTML or text) with a random key, andtermwright lab …on the command line, e.g.termwright lab traffic start voiceortermwright lab traffic report --csv. Its changes are not asked first and show as API in Events. The connector got the new stream fields andset_port. -
Emulated lab: right-clicking a switch closes its hover card, so the menu stands alone.
-
Emulated lab, New Topology page: pick devices by model from a catalog grouped by operating system (search, collapsible groups; a click adds one more) - Termwright picks the image and writes each switch's persona (model, serial and MAC in the model's format) before the first boot; the software version, count, names and role are per row. Cabling: Leaf-spine (the core-capable models are spines by themselves, the others leaves; every leaf to every spine), Ring, Full mesh, Star (around the hub), Line or No links, with 1, 2 or 4 cables per pair (two or more as a LACP or static LAG on both ends), a live preview, what it needs and where it runs. The switches are laid out by the shape.
-
Emulated lab, one canvas toolbar at the top: the floating bar at the bottom and the view bar are one bar under the tabs, with icons and short labels; More (⋯) switches it to icons only or icons and text, and has the mini-map (off now; the Map button is gone), Arrange by Host and how hosts are shown.
-
Emulated lab: an image's ready text can name several signs of a finished boot, separated by
|(any of them counts) - the device pack 1.7.1 counts VOSS as booted at itsLogin:prompt too, so a VOSS switch whose console never prints "Boot sequence successful" no longer stays booting. -
Emulated lab, capture the management port: right-click a running switch → Capture the Management Port records its management traffic (SSH, SNMP, DHCP, its cloud connection…) both ways, in the Captures tab like a cable's; the connector's lab_capture takes port
mgmttoo. -
Emulated lab, light theme: the Slate tiles, Steel isometric and Rack front drawings have light versions, and Draw As opens on the background of the theme.
-
Emulated lab, Captures, packet details: click a packet for its fields (Ethernet, VLAN, ARP, IPv4/IPv6, ICMP, UDP, TCP, LLDP TLVs, STP BPDU, LACP actor/partner, EDP) and its bytes in hex with ASCII under the list; clicking a field marks its bytes, clicking a byte finds its field. Tick packets (or Shift-/⌘-click) for Copy Selected and Save Selected as .pcap; double-click (or Open Packet) opens one in a window with ◀ ▶ through the capture, Copy as Text, Copy as Hex and Save as .pcap.
-
Emulated lab, faster: the lab's view no longer reads every image's password from the system keychain on each refresh (with many images that took seconds after every change, such as a new cable).
-
Emulated lab, Add Link in two clicks: each click takes the switch's next free port and the cable is drawn at once; Change Port (or Alt-click) picks the port instead. Cables of a LAG reach the switch drawing at both ends (there was a gap at one end in some looks).
-
Emulated lab, Grid: a Grid button beside Ports shows grid lines, and switches you move snap to them.
-
Toolbar: the Emulated lab button counts running switches - every host and topology, the ones in the background too (blue), or the switches that need a look - booting, failed, a problem (yellow). Hover lists them by host and topology.
-
Emulated lab: a device pack can change an image's default network card - images still on the pack's old card take the new one once, a card chosen by the user stays. (The device pack moves EXOS-VM to rtl8139, as GNS3 runs it: on QEMU's e1000 cards its LAGs flapped.)
-
Emulated lab: Configure → Edit in Images… opens Images on that switch's own image (it opened on the first one).
-
Emulated lab, Images: Port MAC addresses - the data ports' network cards take the switch's base MAC plus the port number (as before), or a range of their own apart from the switch's addresses, as GNS3 gives them.
-
Emulated lab, topologies side by side: opening another topology (or a new one) no longer stops the open one's switches - they go on running in the background. The topology list shows each one's running switches and hosts; Hosts shows what runs on each host, by topology, with Open. A new, duplicated or imported topology gets its own folder, ports and bridge network, and its switches skip management addresses the other topologies use, so several can run at once.
-
Emulated lab, boot failures in Events: an image can name console lines that mean its boot failed (the device pack does for EXOS: a persona file it rejects, its fail-safe configuration, a reboot by its process manager, an illegal instruction, a kernel panic); Termwright puts them in Events as FAIL instead of showing "booting" for ever - an illegal instruction with a hint about the CPU model.
-
Emulated lab, QEMU command line: each start writes the exact command QEMU runs a switch with (read from the host) to the emulator log at info level; right-click a switch → QEMU Command Line… shows it and copies it. A switch's CPU is taken from that command, so an image's own
-cpu(More QEMU options) also holds on a live move. -
Emulated lab, Images: the Downloading % button moves on with the download (it showed the percentage of when the image was opened).
-
Emulated lab, link check: right-click a running switch → Check Its Links (30 s) turns on QEMU's own trace of its e1000 or e1000e network cards for 30 seconds and says what happened - whether QEMU took links down (the switch restarted auto-negotiation), the switch turned a receiver off, or frames were dropped (link down, multicast filter, overrun). It tells a problem of the emulation from one inside the switch's software.
-
Emulated lab: e1000-82545em in the image NIC choices.
-
Emulated lab, clouds can be selected (click) and have a right-click menu: Add Link to It, take a port off, Remove Cloud (Delete also removes a selected one; the ports on it come off the host's network at their next restart). Test ports have a right-click menu too (Stream from Here, Remove Test Port).
-
Emulated lab, Captures sum up what a capture holds per protocol and sender (LLDP, LACP, STP, EDP… first): one look shows which end sends which control protocol; click one to filter.
-
Emulated lab, traffic generator: put a test port on any free switch port (right-click a switch → Add Test Port…, or Traffic → + Test Port) - a sender and counter on the host where the switch runs, no VM; it has a MAC and an address and answers ARP and ping there. Streams between test ports (or to every other one, or to an address): UDP, ICMP, TCP SYN, ARP or raw L2, with a VLAN tag, frame size, packets per second and seconds or a count; each passes when its loss (and average latency, when both ends are on one host) stays within its limits. The Traffic tab shows sent, received, loss, latency and PASS/FAIL live; Report… saves HTML or CSV; the connector has lab_traffic for AI agents (add ports and streams, start, stop, report). Ports without a cable now have a socket a test port can use (a switch started before this needs one restart for its test ports).
-
Emulated lab, Captures: delete a stopped capture (✕ or Delete, also Delete All Stopped) - its pcap file on the host goes too; two captures of one port each show their own packets; the vendor discovery protocol (EDP) is named as such.
-
Emulated lab, boot menus: a menu that takes one key gets the key alone (Enter only if the menu then still waits); the Enter that used to follow stayed in the buffer and opened the next stop of the boot. The device pack's boot answers and facts patterns now follow the pack as it improves them.
-
Emulated lab, hover card: the state (● running) sits at the top right, apart from the name; Management no longer says "planned, not on the switch yet" for a switch that took its address before Termwright restarted or that resumed from a fast start (the address it took is kept); the serial number read from the switch follows the device pack's newest pattern (an image keeps up with the pack's).
-
Emulated lab, Events has columns: time, host, switch, Level (OK in green, INFO, WARN in yellow, FAIL in red - the message in the same colour, no more boxed rows) and Trigger - USER (you, in Termwright), AI (an agent through the connector) or SYSTEM (Termwright on its own: watching the switches, retrying a copy). Copy includes both.
-
Emulated lab, QEMU versions: a live move keeps the machine type the switch runs as (QEMU's default is each version's and each distribution's own), so it moves live to a host with the same or a newer QEMU; to an older QEMU it is refused with the reason, and a move with its disk to another version boots it there instead of resuming a state that would not load. On the Hosts warns when the hosts' QEMU versions differ, and a host check flags a QEMU older than 4.2.
-
Emulated lab, packet capture: right-click a cable → Capture Packets records it both ways at one end while the switches keep running (QEMU writes a pcap file on the host; the cable turns red and dashed). The Captures tab lists the captures and shows each one's packets live - number, time, source, destination, protocol (Ethernet, VLAN, ARP, IPv4/IPv6, ICMP, TCP, UDP, DHCP, DNS, LLDP, STP, LACP, OSPF, BGP, VRRP…), length and info - with a filter, Stop, Save .pcap… and Wireshark (opens it in Wireshark on this computer, live while it captures). The connector has lab_capture (start, stop, read with a filter, list) for AI agents.
-
Emulated lab, the hover card moves to the left of a switch near the canvas's right edge instead of running off it.
-
Emulated lab, the software version and serial number come from the switch: after it boots, Termwright reads them on its CLI (an image profile's facts commands - EXOS: show version, IMG:; Fabric Engine: show software detail and show sys-info) and shows them on the canvas, in the hover card and the side panel (the serial Termwright gave shows until the switch reported its own). The firmware was empty before. Serial numbers in a persona file can be made SIMxxxx-xxxx, the only form some images take (device pack 1.5.7).
-
Emulated lab, the status bar no longer repeats the topology name and the hosts (they are in the header and the Hosts menu).
-
Emulated lab, three new icon styles, neutral with colour only for the status lights: Slate tiles (a slate tile with a white symbol per role), Steel isometric (grey hardware in 3-D with its ports) and Rack front (1U) (the front of the box as it sits in a rack: ears, port blocks, LEDs). Pick them in the tool dock (all switches) or per switch (Draw As…).
-
Emulated lab, Log tab in the bottom pane: the emulator's own log (emulator.log in the lab folder, three files of 5 MB) with everything the lab does, at a level you choose - Errors, Warnings, Info (default), Debug (each command run on a host, how long it took and how it ended, what day 0 typed) or Verbose (also the commands' text and output, secrets removed). Refresh, Copy and Save….
-
Emulated lab, Fabric Engine on the host's network (macvtap): it answers on its OOB port with its own MAC (its MAC block + 0x81), not the NIC's, so nothing reached it by SSH; the port now gets the MAC the switch uses (an image profile setting: mgmtMacAlign/mgmtMacOffset, in the device pack 1.5.6), and NAT leases are found by it too.
-
Emulated lab, the hover card shows the switch's type (EXOS/VOSS), model (or "no persona") and firmware; the name under a switch can be dragged away from a cable (double-click puts it back) and has a backdrop so a cable does not run through it.
-
Emulated lab, the Topology menu opens again when no topology has been saved yet (it got nothing to show and stayed shut).
-
Emulated lab, console logins no longer lock a switch out: the console is listened to before Enter wakes it, and a password asked for an empty user name gets one Enter, not three (Fabric Engine counts each as a failed login: "Access denied : length 0 not allowed", then "Lock out for 60 seconds").
-
Emulated lab, Console Log tab in the bottom pane next to the live console: everything the selected switch's console showed since it started, with its Events - Refresh, Copy and Save….
-
Emulated lab, canvas: under each switch its model and software - "5520-24T · EXOS 33.6.1.14" with a persona, "EXOS-VM 32.6.3.126" or "VOSS-VM 9.4.0.0" without - instead of the model twice and a "persona ✓" badge (a persona that failed still shows).
-
Emulated lab, Cloud on the canvas: a host's own network as a node; Add Link from a switch port to it puts that port on the host's network (when the switch starts), drawn as a cable with its port; drag it where you like. Add Cloud in the tool dock adds one before any port is on it.
-
Emulated lab, icon styles: Classic diagram symbols and Line icons are no longer offered (switches drawn with them keep them until changed).
-
Termwright connector (MCP) for AI agents, the emulated lab end to end: lab_images, lab_add_switches, lab_cable (connect or remove a list), lab_power (start, stop, restart, shut down), lab_wait (until booted and configured), lab_console (log in on a switch's console and type configuration or show commands, outputs back), lab_check_cabling (LLDP against the cables), lab_move (live, copy or image) and lab_events, next to show_lab. Each change is approved by the user in Termwright, a whole list at once; outputs have secrets removed.
-
Emulated lab, live moves between hosts with different CPUs: a switch started with its host's own CPU used instructions the other host lacks after a live move, and its programs crashed there (EXOS: the login service, so the address step was refused). With more than one host, switches now start with a CPU every host has (Westmere: Intel since 2010, AMD alike), and a live move of one started with its host's own CPU to a host with another CPU is refused with what to do (restart it once, or Save, Copy and Resume, which then boots it there). A login refused because the switch's login service (AAA) did not answer is tried again for a few minutes, and says so instead of asking for a password.
-
Emulated lab, live move progress: QEMU counts the disk at its full size (only what the switch wrote travels), which made the total jump from the memory's 2 GB to 6 GB; once the disk is across the progress counts the memory alone, and the disk line says it is counted at its size.
-
Emulated lab, a live move cut off earlier (Termwright closed in the middle) left a copy of the switch waiting on the new host, and the next move failed with "could not wait on the new host: up …": that leftover is now stopped first. A process is taken for a switch only when it is QEMU (a pid file left from before a host restarted could name another process). Live moves also work with newer QEMU, which dropped the -d option of migrate.
-
Emulated lab, live moves finish: no bandwidth cap, a switch that writes its memory faster than it goes over is slowed down until it can finish (auto-converge), and the last pause may take up to a second. The progress counts its disk too (QEMU up to 9.0 sends it with the memory) - "99 % · about 2 s left" for minutes was the memory done and the disk still going - and says when the switch is being slowed.
-
Emulated lab, a live move to a host on another network works: the switch goes across running, then gets an address of the new host's network typed on its console (the old one removed first) and its SSH session follows - it keeps running throughout. Before, it was refused with "a live move keeps it".
-
Emulated lab, KVM turned on at a host shows at once: a host without KVM is looked at again each minute, Move to Host looks at both hosts' /dev/kvm before it offers anything (Live and the "no KVM" warning were from the last check, and from the topology's hosts only), and the Events say "KVM is on now on …".
-
Emulated lab, Set Up on a host whose user has only Termwright's network commands without a password asked for no password and stopped with "Process exited with status 1" and an empty log: the host check now tells that apart from sudo for anything, Set Up asks for the sudo password then, and sudo's own messages reach the log.
-
Emulated lab, Remove Host: one tick, Everything Termwright put on this host, chooses all of it (images, switch disks, what Set Up did); each can still be chosen alone. Nothing is deleted unless chosen.
-
Emulated lab, KVM is chosen when a switch starts, from what the host has at that moment: a host where KVM was turned on (or off) since it was checked runs its switches right away, without Check Again, and its host entry follows. A start in software emulation says so in the Events.
-
Emulated lab, Hosts: Set Up Again… is always there for a host that has everything, to run every step again (safe to repeat, for example after something changed on the host). Set Up now writes down what it adds (packages it installed, its sudo rule, the user put in the kvm group) in /var/lib/termwright/setup.
-
Emulated lab, Remove Host… first reads the host and asks what to do there: delete its images (with their size and folder), delete the switches' own disks, remove Set Up's sudo rule, take the user out of the kvm group, uninstall the packages Set Up installed (only those it wrote down) - each one a choice, with the sudo password when needed, and a last confirmation before anything is deleted. "Remove, Keep Everything There" only forgets the host; a host that cannot be reached can only be forgotten. Refused while switches of the open topology run there.
-
Emulated lab, Hosts → Set Up: the Set Up… button is next to Check Again (no scrolling down), its panel opens at the top with the steps in one row and the log across the full width, and What the host has ticks each item as its step is done; at the end the host is checked again and its rows show the result at once (no closing and opening again).
-
Emulated lab, hosts without KVM: a switch starting there says so in the Events (software emulation, 5 to 10 times slower, and how to turn on nested virtualization); moving a switch from a KVM host to one without says so in the move dialog. An image profile can give such hosts an Intel CPU model (tcgCpu) for images that do not recognise QEMU's default CPU.
-
Emulated lab, hosts without password-less sudo: Set Up (with the sudo password) can now let the host's user run only the network commands the switches need (ip, iptables, sysctl, virsh, arping, chown of /dev/tap*) without a password, in /etc/sudoers.d/termwright (checked with visudo first). Then switches on that host can use a lab bridge, NAT or the host's network instead of only port forwarding. The host check counts such a rule as sudo.
-
Emulated lab, Images → On the Hosts (also Images on the Hosts… in the Hosts menu): images down the side, hosts across the top, read fresh from each host. Each cell says whether the host has the image, is copying it (with speed and time left), waits, failed (why, Retry), has another file of that name (Replace) or would run low on disk. Make All Hosts the Same copies every image to every host (from this computer, or from a host that has it), with Copy Here, Copy to All Hosts and Copy the Missing Ones Here for less. Keep all hosts the same copies an image to every host once it is downloaded or copied to one; Keep 20 % disk free holds back a copy that would leave less than a fifth of a host's disk free (Copy Anyway). Click a host for its panel: address, QEMU, CPU and load, memory and disk (total and free), where its images are (Change… moves them to another folder, for example a bigger disk, and leaves a link), its checks with Set Up buttons, and which other hosts it reaches - UDP for cables and TCP for live moves, both ways (Check Between All Hosts); Open the Ports… shows the firewall commands to run.
-
Emulated lab, a switch that does not start says why: a port QEMU could not open (a macvtap port on a host without sudo) or QEMU's own log, instead of an empty reason, with what to do. A move to a host whose management needs sudo it does not have is refused before anything is copied, and a moved switch that did not start there no longer says "it resumed there".
-
Emulated lab, moves show their progress: each copy (the image, the booted image, its disk) and a live move's memory show how far they are, how much of how much, the speed and the time left - in the switch's state, the Inventory and on the canvas (a bar under the switch on its way).
-
Emulated lab, Move to Host between networks: a switch whose address is on one host's network can now move to a host on another network. It gets an address of the new host (a host new to the topology whose network is another gets its own lab bridge - change it in Hosts), the address is typed when it starts there and its SSH session follows; it boots there instead of resuming, as its old address would not work. A live move keeps the address, so it says to use Save, Copy and Resume. Only a move from a fixed address to a host that gives addresses by DHCP, NAT or port forwarding is refused, with what to set in Hosts.
-
Emulated lab, copying an image to a host: Termwright checks the host has room first and says how much is free and needed; a copy that breaks off (the connection dropped) is tried again up to four times and goes on where it stopped; the copy is checked with its checksum on the host before it is used, and a damaged one is removed. Two switches starting at once no longer copy the same image side by side (the second waits and finds it there); before, the second copy could delete the first one's image and leave only its checksum file, so later starts thought the image was there. The image itself is now checked too, and a lone checksum file is removed. A failed copy is in the Events log with the reason in plain words. A switch whose disk cannot be made says why (the image is missing on that host, or damaged - then the next Start copies it again) instead of "disk ".
-
Emulated lab, Events: each line shows the host it happened on (with the host's colour) before the switch; Copy includes it.
-
Emulated lab, Move to Host: Live is offered only when both hosts have KVM or both have none (otherwise the dialog says why and Save, Copy and Resume is the default); the Events log opens when a move starts, says "moving to ..." at once, and a refused move is listed there too, not only in a message that goes away.
-
Emulated lab, Hosts menu: simpler - one card per host with its speed (Fast with KVM, Slow without), how many switches run, CPU and memory bars and its management in short words; the cable check is one line with Check now; the long explanation is in the ? tooltip.
- Simulated lab (Transfer → Labs (Simulated, Emulated)…): switches simulated on this computer, each a small SSH server on 127.0.0.1 with a command line - abbreviations,
?help, Tab, VLANs, ports, IP interfaces, save and reset, ping, traceroute and LLDP between the switches over the cables you draw (not a real operating system: no protocols are run). The switches come from a device pack (one.vmfile: models, port layout, commands, outputs, serial formats) that you load with your consent - it shows the pack’s terms and checksum, is kept on this computer only and can be removed; Termwright carries no models itself. A model’s name sets its ports (…-16MW-8T: 16 MW and 8 T ports). New Lab… chooses models, operating system, counts, names, management network and cabling (leaf-spine, chain, ring); the Lab window shows the switches, the topology (drag, details, show config, cables) and the cables; their sessions are in a Simulated Lab folder with live state and Start / Stop Lab. Serial numbers and MAC addresses can be changed; Command Outputs print pasted real output with each switch’s values filled in. Packs are tested with transcripts. - Emulated lab: real VM images on a Linux KVM host (Lab window → Emulated lab). The vendors’ own VM images (
.qcow2, or the settings from a GNS3 appliance file) boot with QEMU/KVM on a Linux server reached over SSH - the real operating system, protocols and APIs included; nothing of Termwright is installed on the host. KVM Hosts are checked (KVM, QEMU, free vCPU, memory and disk, sudo) and Set Up QEMU/KVM… installs QEMU from the distribution's packages, loads KVM and adds the user to the kvm group (with the sudo password, not saved); each image has a profile (vCPU, RAM, ports, ready text, login, day-0 commands, answers) and is copied to the host once, every switch booting from a copy-on-write disk. New Lab sizes the lab against the host, with management by port forwarding (no root) or a bridge (sudo). Each switch shows its boot state; the day-0 commands are typed on the console after the first boot. An SSH session and a console session per switch (through the host as jump host), virtual cables between the VMs that can be pulled and plugged in, Restart, Back to Day 0 and Remove. - Lab serial numbers. New in the Model's Format gives switches serial numbers in the device pack's (or the image's) format; emulated switches take a serial number (as their SMBIOS serial) and a MAC address. Sessions of a lab that are missing are made again when it starts. The emulated CPU column shows the use now (a share of the switch's vCPUs), not the average since boot.
- Emulated lab: management on the host's network. A switch's first port joins the host's bridge (or its network port through macvtap) and gets its address from your DHCP server; Termwright reads it on the console (the image's Address command) after each boot and points the SSH session there. This is the default when the host has password-less sudo.
- Lab buttons in the title bar (a flask for the simulated lab, a cube for the emulated one). The Session Manager's bottom row is New, Shell, Scan (Shell opens a local shell; Import stays in File and the ⋯ menu), in the start page's colours. An emulated switch no longer falls back to booting when its ready text scrolls out of the console log.
- Emulated lab: templates, personas, resources, host use. The tabs follow the work: KVM Hosts, Images (with versions), Templates, Switches. A template is an image with vCPU, memory and disk, the model it acts as, and where serial numbers and MACs start (they count up per switch). Image profiles can bring a persona file (model, serial numbers, modules, stack members) that Termwright writes into a switch's disk with libguestfs while it is off; switches' resources and personas can be changed later. A bar shows what the lab takes of its host now. Images → Import Profiles (.json) reads image profiles and templates; serial numbers from a list can be just the first one, MACs from a first address.
- Emulated lab: devices instead of forms. New Lab rows pick a device (a model the images can be, or an image as it is) and the software version from the image files of that kind; serial numbers and MACs are filled in in the model's real format (from the device pack) and can be made anew. Device packs can bring the emulator's image profiles (
emulator.json), and an added image file is recognised by its name. Persona errors stay visible (they are no longer cleared by the day-0 step), switches are powered off cleanly before their persona is written, libguestfs errors say what to do, and a host that is itself a VM gets a note about passing the switches' MAC addresses. - Emulated lab: the host's NAT network, and management changed later. A fourth management choice, like GNS3's NAT node: the first port joins libvirt's
virbr0on the host and gets an address from its DHCP server; Termwright reads it from the host's DHCP leases, so any image gets an address without a CLI command (the default on a host that is itself a VM). Set Up installs and starts libvirt's default network. The lab's management is now shown and changed in the bar under the tabs: it can be changed while switches run (they are stopped and started again with it, their configuration kept), and their sessions follow. - Fixed: persona files went into the mount-point folder instead of the partition (for example
/scratchon the root filesystem rather than the/scratchpartition), so the switch booted as the plain image. The file now goes on the partition the image mounts there (its/etc/fstab, or a partition labelled like the folder) and is read back (switches whose file was written the old way get it again on their next start); a profile's check command (for exampleshow version) confirms the model after boot, or shows running · no persona with the reason and where the file was written. - Fixed: day-0 stopped at a first-login questionnaire (
… [y/N/q]:): it is answered withq, the defaults. - Fixed (simulated lab): "no host key" when starting switches after the lab was reloaded (for example after loading a device pack). New simulated labs use
10.99.0.0/16(a lab with the old full/24grows to/16), the "network is full" error shows once and says what to do, and the wizard's buttons no longer wrap. - Lab cabling: switches added to a lab go on from its last switch in a chain or ring; the preview numbers new names past the ones in use, as the lab does, instead of showing the same name twice.
- Lab workspace. The lab buttons at the top right open the lab as a tab, in the style of GNS3: a palette of devices to drag onto the canvas, the switches with live states, model and address, cables with the port at each end (pulled ones red), right-click menus, Add Link (switch, port, switch, port), zoom, the selected switch with its persona and problems, the topology summary, the host's use, and an Events log of everything done to each switch - started, persona written where or why not, booted, day-0, persona checked, address, errors - plus the end of its console. New Lab and Add Switches are a panel beside the canvas, which shows the switches to come; Configure has the device model, stack slots, serial numbers, MAC, the persona's values with a live preview of its file and whether it is in the switch, and resources; Add Link shows the switch's front panel (free and cabled ports); Images and Templates and Hosts (resources, what is installed, management choices, Set Up with its steps and log) are pages of the workspace.
- Emulated lab: GNS3 appliances with downloads, units, hosts to choose from. An imported appliance lists each version with its size, MD5 and download link, and Download All fetches the image files (MD5 checked). Versions come from the appliance; image names like
NOSGNS3.9.4.0.0.qcow2are read as 9.4.0.0 (a number glued to a word is skipped). Memory shows MB and GB, disks GB, the disk type is listed everywhere, and More QEMU options has examples. New Lab shows the KVM hosts as cards with their free vCPU, memory and disk; serial numbers and MACs are filled in (with ↻) in New Lab and templates. A device dropped on an empty lab makes the switch at once when the host is clear. - Fixed: a persona file could still go into an empty mount-point folder when the image mounts the partition from its own start scripts (no fstab entry): an image profile can now name what is at the top of the right filesystem (fsHas), an empty folder on a root filesystem is never taken, and when nothing fits the error lists the disk's filesystems with their contents. Switches whose file was written that way get it again on their next start.
- Lab Events pane: taller by dragging its top edge, Copy and Clear; a failed login or day-0 shows the end of what the console printed, a missing persona what the check command printed, and a login refused while the switch still starts is tried again. Downloads show their progress in the button (green when done) and the folder they go to.
- Emulated lab: public appliances. Images → Public appliances lists open-source routers, firewalls, switch OSes, Linux hosts and a traffic generator (and vendor lab images that need your own account) from the public GNS3 registry; Add puts their versions in the image list, ready for Download All. GNS3 appliance files in the newer registry format (several settings, default login, download pages) are read.
- Emulated lab: more disks, CD-ROMs and BIOS files. An image can boot with a second (third, fourth) disk, a CD-ROM and a BIOS file, as GNS3 appliances describe them; each switch writes to a layer of its own on every disk (raw disks and ISO files work too). The image page lists all its files with Choose… and Download (zipped files are unpacked), and the public appliance list now includes images that need them and cloud images with their cloud-init CD-ROM.
- Simulated lab workspace: Load Device Pack in the toolbar (and on the empty palette).
- Fixed: emulated switches showed every port active although only a few were cabled: ports without a cable (and pulled cables) now have their link down after the switch starts, as in GNS3. On the host's network the host's port is put in promiscuous mode, as GNS3's cloud node does.
- Lab downloads: a queue you can order. The Images page shows what is downloading and what waits, in order, with Next, Earlier, Later, ✕ (stop it, or take it out of the queue) and Stop all. Downloads packed as
.gz,.xzor.bz2(as many registry images are) are unpacked before their MD5 sum is checked. - Public appliances can be removed. An added appliance shows how many of its images are in the list and Remove, which takes them out and deletes the files Termwright downloaded for them (not files you chose yourself, and not while a switch or template uses them).
- SONiC-VS in the public appliances (the community build of the SONiC network OS, versions 202411 and master, login
admin): its ports are namedEthernet0,Ethernet4, … - port names can now count in steps (Ethernet{n*4-4}). - Fixed: an emulated switch's SSH session kept asking for the password. The session used the device type's saved login (meant for your real switches of that kind) or, for an image whose login has no password (EXOS'
admin), had none saved; it now always uses the image's own login and password, an empty one included, and follows when the image's password is changed. In all SSH sessions a refused password is asked for three times in all (not three times per login method), the prompt says the saved or typed password was refused, and the error then says the login was refused. - Fixed: day-0 on the console answered questions that were not asked. A question is now answered only while it waits at the end of the console, so text that went by (the first login's questionnaire, a command's output) no longer types
qat the CLI or the login prompt (an EXOS console then logged itself out and refusedqas user name). Day-0 commands that ask to be confirmed ((y/N)) are answered withy; a log message printed after the prompt no longer makes day-0 wait for a prompt it already had (Enter shows it again);Press <SPACE> to continuepages go on by themselves; a check command that asks a question no answer covers says which question. Every answer is in the Events log. The switches' console sessions log in as the image's user too. - Lab colours as in GNS3: a switch is green when it is up, orange while it starts, red when it is stopped or failed to start. A switch that is up with a problem (day-0, persona) stays green with an orange ! (its problem on hover), and the bar counts them apart from failed switches.
- Emulated lab: the host's network, routed (cloud). A management choice for hosts whose network does not let the switches' own MAC addresses through (a host that is itself a VMware or Hyper-V VM, or on Wi-Fi), where bridged switches only got a link-local address: the switches get fixed addresses on the host's network from a first address you pick, with its gateway; the host routes for them and answers ARP for their addresses (proxy ARP), so they reach the network and are reached from it with any protocol (SSH, HTTP/HTTPS, FTP, TFTP, SCP, image upgrades). Images have Set address commands, typed on the console at each start (also for the lab bridge, which so far gave the switches no address); the device pack 1.4.3 brings them for EXOS and VOSS.
- Lab console pane you can type in. The selected switch's console at the bottom of the lab shows the end of its console; a click (or Type here) connects it there, live, without a tab of its own, and Disconnect closes it (while it is connected, day-0 and checks on that console wait, as the console takes one connection).
- Session options: why Proxy cannot be chosen. A session goes through a jump host or a proxy, not both; the page now says so when a jump host is set (lab sessions use the KVM host as jump host).
- Fixed: changing the management while switches ran could be ignored (the switches kept their NAT addresses although the menu showed the new choice): a status read still running from before the stop said they were up. Termwright now waits for a fresh one, says so when switches still run, and the switches start again either way. Their SSH address shows "through the host" for the routed host network and the lab bridge.
- Lab console pane: it says when the switch's console is open in a tab (a console takes one connection) with Go to the Tab, takes the keyboard when it connects, and Console moves a console from the pane to a tab.
- Fixed: the routed host network (cloud) could not be chosen on a host checked by an older Termwright (its own address was not known yet): the host is checked again, and when a management change fails the menu shows the lab's real management again instead of the choice that was not made.
- Lab reachability check: once a switch has its management address, Termwright tries it from the host (ping, SSH port 22, Telnet port 23) and says in Events what answers - so a session that does not connect tells whether the switch or the way to it is the problem.
- Fixed: switches with a fixed address (routed host network, lab bridge) kept a link-local address. EXOS's first login ends with save your configuration changes to primary.cfg? (y/N), which was left unanswered, so the login, the persona check and the address commands waited until they timed out. Yes/no questions at a login now take their default (the capital letter); day-0 commands are still confirmed with yes.
- Fixed: the lab bridge sent the switches' traffic to an old bridge. Bridges of removed labs or of an earlier management (with the same network) are removed when switches start, and a lab's bridge goes with its last switch.
- Management modes tested end to end: a test bench (
scripts/emu-bench) boots a small switch with DHCP, SSH, Telnet and HTTP in each mode - port forwarding, NAT network, lab bridge, routed host network, bridged host network - and checks it gets its address and answers from the host (and from the network for the host's network). The reachability check in Events now covers HTTP too, waits up to a minute for the switch's services, runs for NAT-network switches as well, and says when the host cannot reach a macvtap switch by design. - Lab sessions straight when they can be: on the routed or bridged host network, once a switch has its address Termwright tries its SSH port from this computer; when it answers, the switch's SSH session connects without the jump host (else through the host, as before), and Events says which.
- Fixed: a lab bridge kept the address of an earlier management (the routed host network after the lab bridge, or the other way round), so the host spoke to the switches from the wrong address; the bridge now has only the address of the current one.
- Emulated lab: the bridged host network with fixed addresses. For a network without DHCP, The host's network (bridged) asks whether the switches get their addresses from a DHCP server or as fixed addresses from a first address on (with the gateway); they are typed on the console at each start, and the switches are reached straight from the network.
- Session options: turn the jump host or the proxy off. Each has its own switch on the Jump Host / Proxy page and is turned off without being lost: the connection does not use it, it stays shown (greyed) and works again when it is turned on.
- Emulated lab: fixed addresses that another device uses are found. When the management takes fixed addresses (routed or bridged host network), Termwright asks the network for them while the switches are off and names the ones another device answers on (with its MAC address), so a session does not end up at that device.
- Lab switches in the Session Manager: right-click a switch's session (or its console) to Start, Stop or Restart the switch. Fixed: a starting emulated switch showed a large yellow oval instead of a dot.
- Fixed: a new fixed address was not taken by EXOS switches that had one already (EXOS refuses a second address on the Mgmt VLAN): device pack 1.4.5 removes the old one first, and Termwright now asks the switch which address it has after typing it - when it is not the one given, Events says so (with what the switch showed) instead of pointing the session at it.
- The active tab stands out: tinted with the accent colour, an accent line under it and its name in bold.
- Lab topology: every cable is shown. Several cables between the same two switches run side by side, fanning out from each switch with their own ports (before, they were drawn on top of each other and only one could be seen).
- LAGs in the topology. The cables of a LAG are one thicker track with a ring at each switch that has the LAG (static or LACP, with its ports); a LAG set on one switch only has one ring there, its cables fan out to single ports on the other switch and it is shown in orange with the reason on hover. LAGs are read from the switches after they start (the image's LAG command:
show sharingon EXOS,show mlton VOSS; Read Its LAGs on a switch) and can be set in Termwright on a cable's menu (LACP or static, on both switches or one); what a switch reports wins. - Lab: hover a switch for where it runs, its console (host and port), management address and port, CPU and memory it uses now, serial number, MAC and cables. Inventory (a tab of the lab's bottom pane) lists every switch under its host with its state, console, management, CPU and memory, and the host's CPU and memory; Copy gives it as CSV. The memory a switch uses now is read from the host.
- Teams: Reinstall or Update… for owners, also when the team server runs this Termwright's version: install the nightly build (the newest build of main, with fixes before they are released) or the latest release on the team server, the same safe way as an update (the old container comes back if the new one does not answer).
- Fixed: no check box on the team admin page could be ticked once a theme (Light or Dark) had been picked on it: the theme's click handler also landed on the page itself and drew the page again on every click, before the box could count. Update the team server to get it.
- Team admin page, Discover: a click anywhere on a result row ticks it, and the selection is kept by the page itself, so a box can no longer be left out of step when the page is drawn again during the click. (A team server older than 0.34.1 still has the earlier page: update it with Options → Teams → Update Server….)
- Emulated lab: named topologies. The Topology chip in the lab's bar saves the lab under a name (and keeps it there as it changes), opens a saved one - also with all its switches started at once (Open and Start All) - starts a new one, and deletes old ones. Switches that run are stopped first (after asking); a new topology's sessions get a folder of their own.
- Emulated lab on several hosts. One topology can run on several KVM hosts: Add Switches picks any checked host (the ones already in the lab are marked), a switch's menu has Move to Host (it starts there from its image), and cables between switches on different hosts are UDP tunnels between the hosts (dashed, purple; the hosts must reach each other on UDP 20000-29999). Each host is an area round its switches on the canvas with its address, management and vCPU/memory; each host keeps its own management (the Management chip has a host menu); the side panel shows a card per host; the Inventory lists the switches under their host with console, management, CPU and memory; the switch tooltip names its host.
- Emulated lab: a prompt hidden by log messages after the login. Fabric Engine prints ZTP+ discovery failed right after the login, which hid the prompt and failed the day-0 configuration (no prompt after the password); Termwright now presses Enter to show the prompt again.
- Emulated lab: the factory password while none is set. A device pack can give an image its factory password (Fabric Engine), used for the console login and the sessions until you set one; a login refused without any password now says to set the image's password.
- Telnet: Enter is a bare CR in binary mode. Once binary was agreed (a QEMU serial console asks for it) Enter no longer sends CR NUL: the NUL reached Fabric Engine as a character, so a typed password had one character too many and an empty login followed. The emulated lab also waits out a console lock-out ("Lock out for 60 seconds") before logging in again, does not retry a login refused without a password, and existing images get the pack's factory password and ready text.
- Emulated lab: a first login that asks for a new password. A console answer can type the image's password (
{password}), so a switch that must change its default password at the first login (Fabric Engine: Enter the New password) gets it twice and carries on; the events say "the image's password" instead of the password. - Emulated lab: running switches no longer stay "booting". A switch is ready when its console log since the start shows its ready text (login:), not only the end of it: after a login and some commands, or after Termwright was restarted, the end no longer showed it and the switch stayed orange with booting though it ran.
- Emulated lab: Fast start. Stop saves each running switch's state into its disk (a QEMU snapshot) and Start resumes it in seconds - no boot, no day 0, its configuration, address and sessions as they were. Shut Down on a switch's menu (and Reload) stops it without saving, so it boots next time. A changed persona, management, memory, CPUs, disk size or image makes it boot again, and the old state is dropped; a state that does not load (another QEMU) boots instead. Fast start in the lab's bar turns it off. The saved state takes the switch's memory size on the host's disk.
- Emulated lab: Fabric Engine's "--More-- (q = quit)" is recognised as a page prompt; day 0 and the LAG check waited on it until the switch was stopped.
- Emulated lab: a persona that did not take says why. The check lists the persona file's folder on the switch: the file there means the switch read it and kept its first model (some images take a persona only on their first boot - Back to Day 0), no file means it went to another filesystem of the disk. The check's output drops the banner lines.
- Emulated lab: the persona badge says what is known. ✓ only once the switch showed the model on its CLI (kept across restarts of Termwright); persona ? while the file is written but not checked yet; no persona with the reason when the check found another model.
- Emulated lab: Back to Day 0 writes the persona again. The new disk got no persona file (Termwright still counted the old one as written), so the switch booted as the plain image. Reset also forgets the saved state and the old persona check. Session editor: the Jump host and Proxy switches show off while none is chosen.
- The AI agent and the emulated lab. A new agent tool, Emulated lab (on by default), gives the AI the lab as it is: the switches with model, software, state, KVM host, management address, console and the saved sessions to open, the cables with their LAGs and tunnels, the KVM hosts with CPU and memory, and recent problems. The agent can start, stop and shut down switches, add and remove cables, pull and plug links, mark LAGs and read them from the switches (
lab ...lines, asked like configuration changes; read-only mode refuses them), and configures a switch through its saved session. The MCP server has show_lab for outside AI tools. - Emulated lab: the topology configures the switches, and checks the cabling. A LAG marked on a cable has Set the LAG on the Switches… on the cable's menu: it shows the commands for each switch (from its device profile:
lagSetper mode,lagDel), types them on the consoles and reads the LAGs back; Remove the LAG from the Switches… does the opposite for a LAG the switch has. Check Cabling in the lab's bar asks the running switches for their LLDP neighbours (neighborCmd) and marks every cable as drawn, wrong (it sees another switch) or none; the cable's tooltip and the Events say which. The AI agent haslab setlag,lab unsetlagandlab checkcabling, and sees the check's result on each cable. - Emulated lab: device drawings in five styles. Devices are drawn as Classic 3-D (the blue boxes and cylinders of network diagrams, the default), Classic diagram symbols (round router, switch with crossed arrows, brick firewall), Isometric hardware (3-D hardware with port lights), Line icons or Front panels - all Termwright's own drawings. The style is chosen on the empty lab, or with the style menu beside the zoom buttons; Draw As on a switch's menu picks its kind (access switch, core, router, firewall, AP, server, KVM host, cloud) and its style, for every switch of that model in the topology (new ones too). The choices are kept with the topology and in backups.
- Emulated lab: topologies across hosts, copied and shared. Opening a topology whose hosts are not yours asks where each host's switches go (your checked hosts with their vCPU, free memory and disk); switches on another host start there from their images with its best management. A stopped switch dragged into another host's area moves there. The side panel has Check UDP Between Hosts (each host sends UDP to each other one, as the cables between hosts do) and Add a Host to this Topology. Saved topologies can be duplicated (a lab of its own), renamed, exported to a .twlab file and imported; the list shows all the hosts a topology uses.
- Emulated lab: a data port on the host's network. Port on the Host's Network… on a switch's menu puts one of its data ports on the KVM host's network, like a cloud node on that port (a tap on the host's bridge, or macvtap on its network port); the switch shows ⇡ with the port, and the AI agent has
lab uplink. It applies when the switch starts. - Emulated lab: one boot for a new switch. When a switch powers off after its first boot to take its persona, Termwright keeps that disk as the image's booted copy on the host; the next switches from that image start on it, get their persona before their first boot and boot once. The Fast start tooltip says about how much disk the saved states take.
- Emulated lab: templates are device types. The device list on the left has a Templates group: a template (an image version with its resources, model and module) is dragged onto the canvas like any device.
- Emulated lab: move a switch to another host with what it has. Move to Host (or dragging it into another host's area) now offers Live (No Downtime) for a running switch: QEMU's live migration copies its memory and disks to the other host while it keeps running, then the switch carries on there (a ping or two may be lost); its cables follow through UDP relays. Save, Copy and Resume (Copy Its Disk when stopped) saves it, copies its disk through this computer and resumes it there, keeping its configuration and day 0. From Its Image boots it there from day 0, as before.
- Fixed: a switch moved to a host without KVM did not start ("Could not access KVM kernel module"): each host now uses its own acceleration, and a live move between a KVM host and one without it is refused (use Save, Copy and Resume).
- Emulated lab: Boxes icon style, the new default (the first lab look: a rounded box with a small faceplate), and cables now run to the edge of each drawing in every style instead of stopping short of it.
- Fixed: switches did not start when an old cable relay was cleaned up (the clean-up matched its own start script).
- Emulated lab: short problems in the side panel. A persona problem shows in three lines at most, in plain words (The switch did not take the persona: it does not show 5720-48MW); the whole text is in its tooltip and Details in Events › opens the Events log.
- Emulated lab: the Events log is kept (its last 300 entries) when Termwright starts again or is updated; Clear empties it.
- Emulated lab: a calmer workspace that never reflows. One header row: the topology, page tabs (Canvas, Hosts, Images, Templates, with counts), a health bar with the switch counts, Management (one menu for every host: each host's management side by side, with KVM shown) and Start/Stop All. The canvas tools float at the bottom of the canvas (Add Switches, Add Link, Restart the selected switch, Console All, Check Cabling, zoom and the drawing style). A status bar at the bottom shows the topology, each host with its KVM and management, what is happening now (a switch booting or moving, and how many more), downloads, images, vCPU and memory in use, and Fast start. On a narrow window the header drops words, never a row.
- Emulated lab: a move shows on the canvas. As soon as a switch starts moving to another host, its new copy is drawn among that host's switches, dashed, with the step it is at (copying its disk…) and a moving dashed line from the switch; when the move is done the switch takes that place (a failed move leaves it where it was, the reason in the Events).
- Emulated lab: Save Console Log… beside Console, SSH and Configure (and in the switch's menu): everything its console showed since it started, from its log on the host, with its Events, saved to a file - to keep, or to send when a boot or login went wrong.
- The host list over the canvas (Hosts: Tags) folds away with a click on its title.
- Emulated lab: Draw As in a window of its own. The switch menu has one Draw As… item; the window shows what the device can be drawn as (access switch, core, router, firewall, AP, server, KVM host, cloud) and the six styles, large, with a Dark / Light preview.
- Emulated lab: nothing selected. A click on the empty canvas, Esc, or ✕ beside the switch's name in the side panel unselects it: the side panel shows the whole topology again.
- Emulated lab: moves and copies show in the header. While a switch moves to another host (or an image is copied, a state saved or resumed, images downloaded) the header shows it with its step (⇄ sw-1: copying its disk…, +N for more); a click opens the Events. In a narrow window the header keeps the topology's name and drops the word Topology.
- Fixed: SSH was not on after day 0 with fixed management addresses: the address was set after the day-0 commands (enable ssh2, save configuration). The address now comes first, then day 0, so its commands find the management port configured and the saved configuration has it.
- Emulated lab: no more waiting after a click. Changing an icon, clearing the Events, drawing a cable and the like waited for the switches' state to be read from every host over SSH (seconds with a slow host); the state is now read in the background and the lab answers at once.
- Fixed: day 0 on a switch with a first-login questionnaire never got typed: its
[y/N/q]questions were answered with q, which ends the session (the switch then asks to save and to quit it). They now take their defaults (Enter), also for images added with the old answer; a session that was logged out anyway is logged in again. - Emulated lab: a planned address is shown as planned. With fixed management addresses the side panel and Inventory say planned, not on the switch yet until day 0 has set the address on the switch and the switch shows it. A failed day 0 offers Type Day 0 Again next to the problem (the disk and configuration stay).
- Fixed: day 0 could log itself out after a first-login questionnaire: the switch then asked Are you sure you want to quit this shell session? (Y/n) and the default (yes) was taken. A question whose yes would end the session or the switch's run (quit, exit, log out, reboot, shut down) is now answered no.
- Fixed: day 0 typed the login while the switch's login was not ready yet (
(pending-AAA) login:while its authentication service starts): Termwright now leaves such a prompt alone, types Enter every few seconds until the plain prompt comes, then logs in; an image profile can name more such prompts (notReady). - Emulated lab: a switch's problem in the side panel is three lines at most (the whole text in its tooltip), with Details in Events ›, as for persona problems.
…and more - the full list is in CHANGELOG.md.
Full changelog: v0.34.1...v0.36.0
macOS: the app is ad-hoc signed, not notarized - after installing run xattr -dr com.apple.quarantine /Applications/Termwright.app.