Repository navigation
Releases
v0.38.0
One Memory, one Appearance
Memory is one set of plain notes you can review, Appearance is one page, and there are six new backgrounds.
Highlights
- One Memory: Plain notes per folder or global that threads read each turn. Notes a thread adds after reading untrusted content wait in a Proposed inbox until you accept them, and secrets are never saved.
- Appearance in one place: Settings > Appearance is one page, with six new backgrounds: Dither flow, Contours, Glyph contours, Fish school, Plankton and Ink.
- One width everywhere: Agents, Skills, Memory, Computers and Extensions line up on one page width.
0 memory export --to AGENTS.mdwrites your notes as a file you can review.
Added
- Memory is one set of plain notes, per folder or global, that threads read at the start of each turn. There is one Memory page and one
0 memorycommand:list,add,remove,clear,export,importandproposed. - A thread that has read untrusted content can only propose notes. They wait in a Proposed inbox for you to accept, and are not used until you do.
- Secrets and hidden characters are never saved as notes. 0 saves at most 5 notes per thread and 20 per day.
- Worktrees and clones of one repository share notes.
0 memory export --to AGENTS.mdwrites the notes you wrote or accepted as a file you can review and commit.
Changed
- Settings > Appearance is one page. Mode, Theme, Accent, Background, Text and layout, and Motion each show what they do.
- Theme cards are small copies of the app in that theme, and each Background tile is a real frame of that background.
- Theme and mode live in one place. The profile menu has one Appearance item and the command palette has fewer entries.
- Six new backgrounds replace the old eight: Dither flow (still the default), Contours, Glyph contours, Fish school, Plankton and Ink. A saved choice of a removed background falls back to Dither.
- All pages share one width. Agents, Skills, Memory, Computers, Extensions and the rest line up, and Settings sections are narrower.
0 findings update --status false-positiveremembers why as anavoidnote, and onlyavoidnotes you wrote or accepted reach the scan verifier.ZERO_DISABLE_HUNT_MEMORYstill turns Memory off, next to the newZERO_DISABLE_MEMORY.- The first time Memory opens, the older stores are imported into notes once. Nothing is deleted, and the report is in
~/.0/memory/import-report.json. - Memory is not available inside an isolated workbench for now.
Removed
- The older memory stores: hunt memory, craft memory, triage memories as a separate store, and lesson capture with the lesson pages. They are imported once into Memory and left untouched on disk.
- The
remember_codebasetool and theCYBERGYM_MEMORY_DBoption. - The two learning-comparison bench scripts, which measured the old hunt memory.
Fixed
- The new-thread skeleton race that made CI flaky is now a deterministic test.
- The What's new dialog no longer shows an old screenshot.
- Refusing an approval run by a thread says "This command was run by a thread, not by you. Run it yourself in a terminal to approve." instead of naming an internal variable.
- A command run from another folder than the console says "This console is for . Run the command there, or add --db-path ."
- The Extensions page scope row says On with no switch only when a scope is defined.
- The docs pin the current release, and
node scripts/bump-docs-pins.mjs --checkfails when a pin lags. 0 usage --bytakesmodelorthread, and leftover mentions of plugins and Sentinels in command help were reworded.
Install
curl -fsSL https://raw.githubusercontent.com/0sec-labs/0security/main/install.sh | bash
0 --versionWindows (experimental): download 0-windows-x64.exe below. Docker: docker run --rm ghcr.io/0sec-labs/0security:latest --help.
Verify downloads against the attached checksums.txt (SHA-256), for example sha256sum -c checksums.txt --ignore-missing.
Full changelog: https://github.com/0sec-labs/0security/blob/v0.38.0/docs/CHANGELOG.md
v0.37.0
Verified means checked
A fix is only called Verified when 0 ran a check for it, the Findings board is back, and the sidebar, Skills page and docs are simpler.
Highlights
- Verified means checked: A skill can declare a check that 0 runs in a container before anything is called Verified. Fixes and threads end with Verified by or Not verified, and the second-Agent check is now called Corroborated.
- Findings board: Switch Findings between List and Board and drag a finding between columns. The Agents list is back in the sidebar, next to a Threads toggle.
- Simpler and steadier: The Skills page is a flat list with search, the docs are 12 pages, and many small errors are fixed, from raw reconnect messages to threads that kept the title New thread.
Added
- A skill can declare a check that the harness runs in a container before anything is called Verified. The Agent does not run it. The check uses a pinned image with the network off after setup, a timeout, and success as exit 0, a pattern or no worse than before.
- Fixes and threads now end with "Verified by ..." when the check passed or "Not verified: ..." with the reason when it did not.
- A built-in
workflow-hardeningskill checks GitHub workflows with zizmor and actionlint. Run it with0 threads start --skill workflow-hardening. 0 scan fix --verify-image <image@sha256:...>runs--test-commandin a container with the network off. Without it the command runs on this machine and is never labelled Verified.- The second-Agent check is now called Corroborated. Verified is kept for checks the harness ran itself.
- The Findings board is back, with a List | Board toggle and drag and drop between columns.
Changed
- The sidebar has an Agents | Threads toggle with Agent and thread rows, a shared header with Sort by and Density, and the Agents list is back on the left.
- The Skills page is a flat two-column list with a tile for each skill and search at the top right.
- Extensions connect 0 to your tools and services: GitHub, MCP servers and rules like scope. The pages and help text now say so.
- Scope shows On, with no switch, while a scope is set. GitHub shows Built in.
- Automations are created and edited inside Agents.
/automationsand its old addresses open Agents. 0 skills addflags work before or after the source, and--yesinstalls without a terminal (CI, scripts). An Agent session still refuses.- The docs went from 49 pages to 12.
- CI went from 14 workflows to 5, with one shared build and a release that waits for a green Main on the tag's commit.
- Skills now live at
/skills;/plugins/skillsredirects there.
Removed
- The Reports, Scans and Engagements pages. Findings and Threads carry the same findings, evidence and exports, and the old addresses open Findings. The scan, report and engagement APIs are unchanged.
- The Automations page. Automations live inside Agents.
- Unused web code: dead API client functions, imports and components. A few old scripts and checks are gone too.
Fixed
- Opening a page while the server is down shows "Couldn't load this page. It will retry when the connection returns." instead of a raw error.
- A new thread gets its title after the first answer. It stayed "New thread" before.
- A pull request or scheduled Agent run that fails now says why on the activation, for example an invalid API key.
- Server mode with an explicit port stops with a clear error when the port is busy, instead of silently listening on a random one.
- An extension's own MCP server now starts. It was refused because it runs in the extension's folder.
- A changed or new MCP tool list is no longer approved automatically when commands run without asking.
- A thread can no longer install or approve an extension by clearing the agent-session variable or wrapping the command in
env,sudoornode <script>. - A thread can run or read a file inside an installed extension, so connector scripts get their approved token. State files under
~/.0/extensionsare still protected. - Messages that named a command that does not exist now point at
0 --headless, and a console that belongs to another folder says which folder. 0 findings list --alland--format sarif|json|mdwork as documented. Without IDs the export covers every finding and honours--severity,--category,--statusand--triage.0 findings listand0 agents set --statereject an unknown value and list the valid ones.0 threads list --format textis a table. Agent and scan threads are titledAgent: ...andScan: ..., not Sentinel or Workflow.- …and 8 more in the full changelog.
Install
curl -fsSL https://raw.githubusercontent.com/0sec-labs/0security/main/install.sh | bash
0 --versionWindows (experimental): download 0-windows-x64.exe below. Docker: docker run --rm ghcr.io/0sec-labs/0security:latest --help.
Verify downloads against the attached checksums.txt (SHA-256), for example sha256sum -c checksums.txt --ignore-missing.
Full changelog: https://github.com/0sec-labs/0security/blob/v0.37.0/docs/CHANGELOG.md
v0.36.0
Extensions
Extensions replace plugins. They connect 0 to your tools and services: GitHub, MCP servers and rules like scope. Each is pinned to a commit, with one decision point before every command.
Highlights
- Extensions: Add an extension pinned to a commit with 0 skills add @. A prompt shows what it can do and the diff on update, and the new Extensions page replaces Explore.
- Safer MCP servers and scope: MCP servers live inside extensions and start on demand, every tool is pinned and blocked until you approve a change, and scope is on by default whenever a scope is set.
- Plugins are gone: The old plugin system, its registry and self_extend are removed. Installed plugins are kept disabled and imported once, and GitHub is now a built-in connector.
Added
- Extensions connect 0 to your tools and services: GitHub, MCP servers and rules like scope. Each is pinned to a commit and never updates on its own, and an extension can come with a skill that explains how to use it. Install one with
0 skills add <repo>@<sha>. A prompt shows what it can do and, on update, the diff since the last approved version. 0 skills extensions list|info|enable|disable|update|removemanage extensions from the CLI. A folder that no longer matches its pinned commit refuses to load, and scripts get a scrubbed environment with only the names you approved.- The Extensions page (
/extensions) lists installed and built-in extensions, what each can do, its source at a commit and an on/off switch. Add shows the install prompt and Update shows the diff. - MCP servers now live inside extensions and start on demand. A thread's default tool list carries one line per server and no MCP schema; tool schemas load only when the model asks for a server. Add one from the Extensions page (Add MCP server) or
0 skills extensions mcp. - Skills inside an enabled extension appear on the Skills page, in
list_skillsand inload_skill, labelled with the extension's name. Disabled, refused or tampered extensions provide none. - Connectors are extensions. A connector declares how it connects, what it may read and write, the hosts it may reach and a health check, and the Extensions page shows one Connect card each. GitHub is a built-in connector. See Connect a service.
- A one-time importer carries earlier skills and custom MCP servers into extensions, reports what it could not migrate and never deletes anything.
Changed
- Every command now goes through one decision point (
beforeCommand) covering bash, network targets and the Ask before running commands setting. Any deny wins, then any ask, and a scope denial beats a prompt. - Scope is a built-in extension that is on by default whenever a scope is set (
--scope, a thread, Agent or automation scope, a pull request checkout). With no scope set everything is allowed. This reverses the 0.30 to 0.33 rule that a scope file alone never activates checks. - If the scope extension is off or errors while a scope is set, commands are denied instead of allowed. Projects that already enabled scope keep it enabled.
- Every MCP tool is pinned by a hash of its name, description, schema and annotations. If a server changes, adds or removes a tool, the server stays blocked until you approve it again, and
0 skills extensions info <name>shows Changed, New and Removed. - MCP tool results are treated as untrusted: they are labelled, fenced and capped at 32 KB. Launchers such as
npx,uvxandpipmust pin an exact version, and remote hosts that resolve to private, loopback or metadata addresses are refused. 0 skills add owner/repogoes through the extension prompt (pin it with@<full sha>to install). Catalog ids andowner/repo/pathkeep the earlier skill flow.
Removed
- The old plugin system is gone. Installed plugins are kept on disk, imported once as disabled entries marked "could not migrate" and never deleted. Re-express them as scripts run through bash or as MCP servers.
- Executable plugins: the
plugin.jsruntime, loader, wire protocol and the per-tool approvals for them (first shipped in 0.16.3, review UI in 0.34.0). self_extend, the live harness (agent drivers, plugin views and harness trust) and theallowModelSelfExtensionsetting (0.14.0 to 0.16.3). The model writes a skill or script folder with bash and you install it through the install prompt.- The plugin registry, the hackstore index,
ZERO_REGISTRY_URL, plugin submit (0.14.0), andZERO_REGISTRY_PUBLIC_KEY_FILEwith Ed25519 registry signatures (0.30.0). Signatures and publisher identity are not offered. - The Plugins page, plugin detail page and Explore plugin entries (0.29.0 to 0.32.0), the browser MCP sheets and service plugin adapters (0.29.0), and their
/api/consoleroutes./pluginsredirects to/extensions. - The Explore page and its Plugins and Themes tabs. Themes and backgrounds live in Settings > Appearance, and old Explore links open Extensions.
0 skills runand the0 plugincommands.plugin list|search|info|enable|disable|installandplugin mcp add|list|removestay as hidden aliases that printmoved: use ...; the rest printremoved: ....
Fixed
- A malformed glob in a command no longer fails the protected-paths check.
- Extension folder checks, update diffs, imports and git fetches are stricter, and built-in enablement has one store that is locked and audited.
Security
- An agent can no longer approve its own extensions by flag. Approval commands refuse from an agent session and without a terminal, and commands from an agent that touch approval, enablement or settings files are denied. These are layers.
- Every executing tool (monitor, pty_session, python_exec, run_command and file tools) now goes through
beforeCommand, and extension approvals are sealed. - A connector's stored token reaches only the scripts and MCP servers of the extension that declared and was approved for it, and is masked in what comes back.
- MCP approval prompts show sanitised plain text with an old versus new description and schema diff. Extension server ids are namespaced (
x-), and a dead process or hung call blocks the server.
Install
curl -fsSL https://raw.githubusercontent.com/0sec-labs/0security/main/install.sh | bash
0 --versionWindows (experimental): download 0-windows-x64.exe below. Docker: docker run --rm ghcr.io/0sec-labs/0security:latest --help.
Verify downloads against the attached checksums.txt (SHA-256), for example sha256sum -c checksums.txt --ignore-missing.
Full changelog: https://github.com/0sec-labs/0security/blob/v0.36.0/docs/CHANGELOG.md
v0.35.0
Nine commands, one mode
The CLI is down to nine commands, 0 opens the console, and Threads and Agents have shell access with a small core tool list.
Highlights
- Nine commands: Run 0 to open the console. scan, findings, threads, agents, skills, memory, models, computers and config cover the rest, and old commands still work for one release with a short moved note.
- One mode with shell access: Read-only and Can make changes are gone. Scope, budget, the sandbox and the Ask before running commands setting limit what Threads and Agents can do.
- Faster and steadier: 0 --version starts in about 0.15 seconds, Threads and Agents load a small core tool list, and a long-running console no longer fails every call after heavy use.
Added
- Choose the animated background behind a new thread in Settings > Appearance > Background. Dither stays the default. Nine others are live previews that follow the colour theme and pause when the tab is hidden.
0 threads start --cost-ceiling <usd>stops a thread at a spend limit.--waitnow prints the reply with its tokens and cost, and0 threads show <id> --jsonshows its usage.0 web --headlessserves the API only, so it starts without built dashboard assets. A missing 0verse environment is skipped with one log line instead of stopping the console.
Changed
- The CLI is nine commands:
scan,findings,threads,agents,skills,memory,models,computersandconfig. Every old command and flag still works for one release and prints one dim `moved: use ... - Running
0with no arguments opens the web console. It takes the old0 webflags, for example0 --headlessor0 --port 8080. Automations are underagents --automation, plugins, MCP servers and connectors are underskills, and0 config doctorreplaces0 doctor. scan <target>is the one scan command.--module deep|repair|campaign|deps|deps-fixreplaces the old scan subcommands.findings update,threads send new,agents set,models set,models connectandconfig get|setreplace the older verbs.computers addis now the command name, withcomputers deployas the alias.- There is one mode: Threads and Agents have shell access. Read-only and Can make changes are gone from Threads, Agents, automations, scans, presets, slash commands, the browser app and the CLI.
- Saved Threads and Agents from older versions still load and are not rewritten. An Agent that was read-only is marked as needing review once because its authority widened.
--access,--mode,--autonomyand--tool-accessare accepted and ignored for one release. - Pull request Agents run scoped to the repository's checked-out folder and still skip fork pull requests and drafts by default. The shell gets an allowlisted environment, so other provider keys and tokens are stripped unless you set
ZERO_BASH_ENV_ALLOW. - Threads and Agents now give the model 11 core tools instead of 17 plus a catalog of about 40 more. Specialist tools such as browser, HTTP probes, cloud, proxy and delegation load only when a skill that needs them is loaded. Set
ZERO_TOOLSET=fullfor the previous list. load_skillwith no skill ID lists the available skills. Old tool names keep working for saved threads.0 --versionand0 --helpstart in about 0.15 s instead of 0.72 s because command modules load only when their command runs. They no longer run the startup update check or print the telemetry notice.0 scan evaland0 scan agent-assuremove into plugins after this release, andautomation versionsandautomation rollbackare removed after it (use git history).
Removed
--runtime ollamaand--runtime geminiare removed and fail with a clear error: use an OpenAI-compatible endpoint with--runtime api. Ollama serves one athttp://localhost:11434/v1; see Models and keys.- The legacy
sourceReviewengine is deleted. The oldscan()and package audit paths now live behind the internallegacy-benchdirectory.
Fixed
- A long-running console no longer stops answering after heavy use. Model calls used one shared HTTP/2 connection that the provider could close, after which every later call failed with
fetch faileduntil restart. 0 threads start --folderno longer fails with "Console thread is already processing a turn". The first prompt after a folder change now waits for the change instead of being refused, and no empty thread is left behind.0 threadscommands now fail with "No console running. Start one with: 0 web --headless" and a non-zero exit when no console is running, instead of printing[].
Install
curl -fsSL https://raw.githubusercontent.com/0sec-labs/0security/main/install.sh | bash
0 --versionWindows (experimental): download 0-windows-x64.exe below. Docker: docker run --rm ghcr.io/0sec-labs/0security:latest --help.
Verify downloads against the attached checksums.txt (SHA-256), for example sha256sum -c checksums.txt --ignore-missing.
Full changelog: https://github.com/0sec-labs/0security/blob/v0.35.0/docs/CHANGELOG.md
v0.34.0
Updates that finish themselves
The console can update and restart itself, pull request Agents react to GitHub straight away in server mode, and plugin reviews show which tools changed.
Highlights
- Update and restart in one click: The Update available card downloads, checks and installs the new version, then restarts and reconnects by itself. With Updates set to Automatic it installs in the background once nothing is running.
- Faster pull request Agents: In server mode, GitHub can tell 0 about a new pull request the moment it opens, so an Agent starts reviewing without waiting for the next check.
- See what changed in a plugin: Reviewing a plugin update marks each tool as Changed, New or Removed, and four unused enterprise hooks are gone.
Added
- Update and restart in the console. The Update available card now has an Update and restart button for standalone installs: 0 downloads the release, verifies its checksum, replaces the binary and restarts on the same port and database.
- A running console keeps itself current. With Updates set to Automatic it checks every six hours and installs a verified release in the background when nothing is running, then restarts after 10 idle minutes. Server mode and service-managed consoles show the card instead.
- Server mode accepts GitHub webhooks so pull request Agents react at once.
POST /api/webhooks/githubchecks a signature, accepts onlypull_requestandping, and just asks the poller to check now, so polling stays the default and every review rule is unchanged. - The per-instance webhook secret is generated on first use, stored with mode 0600, shown once, and can be rotated; the old secret keeps working for 5 minutes.
pnpm test:webhook-e2eruns a signed delivery, a replay, a bad signature and an oversize body against a throwaway server-mode console.
Changed
- Reviewing a GitHub plugin update in the web app now marks each tool as Changed, New or Removed against the installed version, using the same per-tool hashes as the install check. Approval is still for the whole plugin.
- The update card reads Update available · 0.34.0 with You have 0.33.0 and a What's new link, instead of repeating the product name. Settings → Updates (Off, Notify me, Automatic) stays the one control, with a plain description.
- Agents is now the main page address:
/agentsis the route and/sentinelsredirects to it, along with the other old addresses. - Surfaces across the console use one corner radius (
rounded-md), and a check keeps new styles from drifting. - Copies older than this release cannot be updated remotely. Run
0 doctor upgrade, or the install one-liner, once. - Internal cleanup only: shared engine types moved to neutral modules, and unreferenced helpers and benchmark tooling were deleted. The Docker publish job now refuses to push any name other than
ghcr.io/0sec-labs/0security.
Removed
- Four unused enterprise hooks that shipped in v0.30.0 are gone: audit event subscriptions for plugins, offline verification of signed licences, support for an external identity provider for team sign-in, and the named-permission policy behind team roles.
Fixed
- A page left open while the console restarts no longer reloads into the browser's error page.
Security
- The GitHub webhook is the only anonymous
/apiroute in server mode, and it is authenticated by its signature. Audit eventswebhook.received,webhook.ignoredandwebhook.rejectedrecord a reason and never the payload, secret or signature.
Install
curl -fsSL https://raw.githubusercontent.com/0sec-labs/0security/main/install.sh | bash
0 --versionWindows (experimental): download 0-windows-x64.exe below. Docker: docker run --rm ghcr.io/0sec-labs/0security:latest --help.
Verify downloads against the attached checksums.txt (SHA-256), for example sha256sum -c checksums.txt --ignore-missing.
Full changelog: https://github.com/0sec-labs/0security/blob/v0.34.0/docs/CHANGELOG.md
v0.33.0
Two kinds of access
Threads, Agents, automations and scans are now Read-only or Can make changes, and asking first is one global setting.
Highlights
- Two kinds of access: Read-only or Can make changes: Agents and threads that asked first now follow one setting, Ask before running commands, which is off by default. Turn it on in Settings to keep confirmations.
- The same choice in the CLI: Use --access read-only or --access changes on 0 agents create, 0 threads start and 0 lab console. The old flags still work.
- Nothing is lost on upgrade: Saved Threads, Agents and automations still load, and the first start shows one note in Settings > Helpers that you can dismiss.
Changed
- Upgrade note: Agents that used Ask first and threads that used the ask-first modes now follow the global Ask before running commands setting, which is off by default, so turn it on (Settings, Helpers, or
0 config set approvals ask) to keep confirmations. - Access has two choices everywhere: Read-only and Can make changes (runs commands, edits files and opens pull requests). Threads, Agents, automations, scans, presets and slash commands all use just these two, replacing the three Agent access levels (Read-only, Can run…
- Whether commands ask first is no longer a per-thread or per-Agent choice. It is only the global Ask before running commands setting, off by default.
- Read-only refuses every command and edit, whatever the scope says. Access is separate from scope, which still means what a scan or Agent may target.
Install
curl -fsSL https://raw.githubusercontent.com/0sec-labs/0security/main/install.sh | bash
0 --versionWindows (experimental): download 0-windows-x64.exe below. Docker: docker run --rm ghcr.io/0sec-labs/0security:latest --help.
Verify downloads against the attached checksums.txt (SHA-256), for example sha256sum -c checksums.txt --ignore-missing.
Full changelog: https://github.com/0sec-labs/0security/blob/v0.33.0/docs/CHANGELOG.md
v0.32.0
Simpler approvals, calmer Threads
Threads run commands without asking by default, slash commands match the CLI, and the sidebar has seven items.
Highlights
- Simpler approvals: Threads and Agents run commands automatically. One setting, Ask before running commands, turns the questions back on. Agents are Read-only, Can run commands or Can open pull requests.
- Slash commands match the CLI: /threads list, /findings list and /config get work like their 0 commands, and the old names still work.
- A calmer Thread: One shimmer for the live work, one row per call, clear messages when a send fails or a reply comes back empty.
- Seven sidebar items: Skills has its own page with Installed and Browse, real brand logos appear on Explore, and the Findings board drops cards where you let go.
Added
- Real brand logos on Explore and Skills rows: each connector mark sits on its brand colour in a 48 px tile (white or black glyph, whichever reads better), and skills show the publisher's avatar with stars and the owner's name. Nothing is fetched from a third-party site.
- A Thread turn that ends without an answer says so. A quiet line under your message reads "The model returned no reply.", "The reply was cut off.", "The provider blocked this reply." or "The connection dropped before the model answered.", with Retry, Continue or Switch model.
- A failed send now says what happened and what to do: your message could not reach the model, the connection dropped, the model took too long, the provider is busy or rate limiting, the model is unavailable, the thread is too long, a folder needs access, or the connection needs…
- Agents have three access levels: Read-only, Can run commands and Can open pull requests. Create one with
0 agents create --access read-only|commands|pull-requests. Existing Agents keep exactly their behavior. - A setting, Ask before running commands (Settings, Helpers;
0 config set approvals ask), also--approvals ask|autoon0 lab consoleand0 threads start|set. - The container image is also published as
:edge, rebuilt nightly from the newest main commit whose checks passed.
Changed
- Chat slash commands are now the CLI commands:
/threads start,/threads list,/findings list,/models default,/config get,/agents list,/doctorand more mirror0 threads startand so on. - The sidebar has seven items and no More menu: New thread, Findings, Automations, Skills, Memory, Computers and Explore. Computers is always listed and no longer appears under Settings.
- Skills has its own sidebar entry with Installed and Browse tabs. The GitHub skills directory moved from Explore to Skills > Browse, and Explore keeps Plugins and Themes. Old
/plugins/explore?tab=skillslinks redirect to Skills > Browse. - Approvals are simpler. Threads and Agents run commands automatically by default and never ask first, unless you turn on Ask before running commands. The old approval wording is gone from the app, CLI help, errors and docs.
- Scope now means one thing: what a scan or Agent may target. The thread Permissions card is the Scope card, its two confirmation checkboxes and "Limited to" text are gone, and your chosen folder is always allowed. Third-party target authorization remains an optional plugin.
- The thread work log is calmer. Only the newest active item shimmers (the running call, else the Thinking row, else the Working label), so a Thread has one shimmer at a time.
- The thread composer dock is tighter and cleaner: the transcript no longer dims its last lines above the composer, there is less empty space under the footer row, and the Scroll to latest pill is smaller and sits clear of the card.
- Hover and focus follow one language across the app: rows, cards and table rows get a light overlay, bordered chips and selects step from button to button-hover, text fields and switches draw a 1px inset focus ring, sidebar thread row actions appear on hover or focus (always on…
- Thread message actions (time, Reply, Copy, More) sit in reserved space above each message instead of covering its first line, with tooltips, a "Copied" confirmation and keyboard and touch access. The code block copy button shows a tooltip too.
- Old installs find and apply updates whatever the repository is called: the latest-release lookup falls back from the API to the release redirect, the repository id and the old name, the installer script falls back between names, and an upgrade replaces the folder of the running…
- The container image is published only as
ghcr.io/0sec-labs/0security;ghcr.io/0sec-labs/0is no longer updated, so switch to the new name. Images are published for releases and for nightly:edge, not for every commit on main. - For contributors: CI runs fewer workflows. Docs-only changes skip test groups, Windows storage and dependency audits run only for the paths they cover, superseded integration runs are cancelled, and Main also runs on release tags. A guard keeps workflow triggers lean.
Removed
- npm distribution is discontinued; install with
install.sh, the release binaries or the Docker image. A check fails CI if npm publishing returns. - The composer options menu (Ask before acting, Act without asking) and the two confirmation checkboxes in thread settings.
- The notice that announced a silently granted local scope in threads.
Fixed
- Findings board: dropping a card lands in the column under the pointer, a card moves out of Accepted and Cancelled at once, Accepted accepts drops, columns are reachable by edge auto-scroll, touch and keyboard (Space, arrows, Space), and refused moves (review stages, a finding…
- Reopening a thread no longer shows calls that have not started as Running.
Install
curl -fsSL https://raw.githubusercontent.com/0sec-labs/0security/main/install.sh | bash
0 --versionWindows (experimental): download 0-windows-x64.exe below. Docker: docker run --rm ghcr.io/0sec-labs/0security:latest --help.
Verify downloads against the attached checksums.txt (SHA-256), for example sha256sum -c checksums.txt --ignore-missing.
Full changelog: https://github.com/0sec-labs/0security/blob/v0.32.0/docs/CHANGELOG.md
v0.31.0
Threads, Agents and a leaner CLI
Missions are now Threads, Sentinels are now Agents, and the CLI has 14 commands.
Highlights
- Threads and Agents: Missions are now Threads and Sentinels are now Agents. The old CLI names still work, and Memory has its name back.
- A leaner CLI: 14 top-level commands now cover threads, agents, automation, models, skills, memory and plugins, with the old names kept as aliases.
- Connections that tell the truth: A model shows Key rejected when its key does not work, and a failed message offers a working model to switch to.
- Run it on a server: An experimental single-node server mode, a Docker image that serves the console, and anonymous usage and crash reports you can turn off.
Added
- The CLI has 14 top-level commands:
web,scan,findings,threads,agents,automation,models,computers,skills,plugin,memory,config,doctorandapi. Old command names keep working as hidden aliases with the same behaviour. - New CLI groups cover what the console does:
threads,agents,automation,findings exportandfindings keys,skills,memory,config getandconfig set,computers list, andmodelswith connections and usage. plugin connectors,plugin mcpandplugin github installmanage plugin connectors, MCP servers and GitHub installs, andconfig audit-logreads the audit log.- Experimental single-node server mode.
0 web --host 0.0.0.0 --server-modeserves the console behind a TLS proxy. It is off by default and local use is unchanged. - Server mode requires sign-in (team accounts, or an access token of 32 or more characters), an explicit
--allowed-hostlist, and--trusted-proxyand--public-urlfor the proxy. - The web console port answers
/healthzand/readyz, and aSIGTERMdrains running work gracefully (--shutdown-timeout). The audit log records console sign-ins, failed sign-ins, throttling and refused hosts. - The Docker image now serves the web console by default, exposes port 48123, has a health check, keeps its state in a
/datavolume, runs undertiniand carries image labels. Release builds also publishvX.Y.ZandX.Ytags. deploy/compose/holds a single-node Docker Compose file, an.env.exampleand an optional Caddy profile with automatic HTTPS.- The Docker image includes the OpenSSH client, so remote computers can be connected over SSH from a container.
0 computers deploy --sshnow says how to installsshwhen it is missing. - Anonymous usage and crash reports are on by default in official release builds, with a notice on first run. Usage holds counts only: tool and error counts, findings by severity, turns, time, cost, version, operating system and a random install ID.
- Reports never hold targets, findings, code, prompts, file paths or error messages. Turn them off with
ZERO_NO_TELEMETRY=1, or0 config set analyticsLevel offand0 config set diagnosticReporting off, or in Settings, Privacy. See Telemetry. - Updating a plugin installed from GitHub says which tools were added, removed or changed. Each tool has its own hash over its name, description, parameters and capabilities, so a changed description is caught too. Plugins installed earlier count as having all tools unreviewed.
- …and 1 more in the full changelog.
Changed
- Missions are now Threads and Sentinels are now Agents, in the console, the CLI and the docs.
0 missionsand0 sentinelsstill work. The Settings page that was called Agents is now Helpers, and "this machine" now says "this computer". - Memory has its name back after 0.30.0 renamed it.
/memoryis the page address again, and/learningredirects to it. - Automation templates no longer talk about a separate place to run, and automations run in the current folder.
- A connection reads Connected only when the provider accepted its key. Connections, Models and Diagnostics show Key rejected or Can't reach with a one-line hint, including for keys set in the environment.
- Choosing Desktop, Documents, Downloads, iCloud Drive or an external drive on macOS waits for the permission prompt instead of accepting the folder after three seconds.
- GitHub Releases are now titled with the bare tag, for example v0.31.0.
Removed
- The Suggestions view in Memory and the Learning section on the Automation page are gone. Memory lists saved memory only.
Fixed
- Pull request Agents now skip fork and draft pull requests by default and record a skipped activation with the reason, instead of reviewing them. A pull request marked ready for review is reviewed once. Closed and merged pull requests are ignored.
- New threads and Home no longer default to a model from a provider whose key was rejected, and a remembered choice from one is skipped.
- An empty thread no longer keeps a model its provider stopped listing. It switches to a usable model and the model button matches. A thread with history says the model is no longer available and waits until you pick another.
- A message that fails with a rejected key names the provider and model and offers Switch to a working model, which resends the message, and Manage connection. Ask 0 to get started shows Connect a model to get started when no model works.
- macOS privacy blocks on protected folders are explained: the picker names the app to allow and offers Open Privacy settings.
- Folder fixes: the chooser no longer opens for a browser on another computer or without a desktop session, closing the popover closes the chooser, and a switch to a symlinked, differently cased or decomposed-unicode path no longer waits 8 seconds.
- Home shows its composer at once, loading shows a thread-shaped skeleton, and loads no longer stick or flicker.
- A browser tab left open across an update reloads itself once instead of showing "Something went wrong".
Security
- Plugin updates report per-tool changes, so a tool whose description changed is visible when you review the update (#367).
Install
curl -fsSL https://raw.githubusercontent.com/0sec-labs/0security/main/install.sh | bash
0 --versionWindows (experimental): download 0-windows-x64.exe below. Docker: docker run --rm ghcr.io/0sec-labs/0security:latest --help.
Verify downloads against the attached checksums.txt (SHA-256), for example sha256sum -c checksums.txt --ignore-missing.
Full changelog: https://github.com/0sec-labs/0security/blob/v0.31.0/docs/CHANGELOG.md
v0.30.0
Simpler navigation, GitHub sign-in
A shorter sidebar, a Models page, a guided GitHub connection and ready-made Automations.
Highlights
- A shorter sidebar: New mission, Findings, Automations and Explore, with Skills and Plugins inside Explore and Lessons under More.
- Connect GitHub your way: Reuse your GitHub CLI login, sign in with a code, or follow a guided token. You see what permissions are missing.
- Eight Automations to start from: Pick a starting point such as a code review or a secrets check. Code reviews use your current project.
- One Models page: Providers and models live together at Models, and a switch hides cost and usage if you prefer.
Added
- Add a GitHub connection card in Settings that reuses an existing
ghlogin, signs in with a device code, or guides you to a fine-grained token with the right permissions prefilled, and shows a "Needs permissions" state when the connected account lacks access. - Add a What's new dialog that opens once after an update, and a sidebar card that reopens it.
- Add eight outcome-named starting points for Automations (code review, deep code review, dependency check, secrets and config check, smart contract review, native code review, web app test and API authorization test), each with a ready-to-use prompt.
- Add a summary-first run dialog for Automations that shows where it runs, the number of steps, the time and cost caps and the model on one line, with the step detail behind Details.
- Add a Show cost and usage switch in Settings that hides cost, token counts and the context meter.
- Add
0 reports exportand ashare-reportoption in the GitHub Action, which write areport.share.jsoncopy without evidence, proof-of-concept steps or local paths; see Reports. - Add audit event subscriptions for plugins: a plugin can receive every audit event after it has been redacted, and a failing subscriber never affects auditing.
- Add offline verification of signed licences for entitlements; no signing key is bundled yet, so no licence is accepted today and nothing changes for open-source use.
- Add Ed25519 verification of plugin registry entries when you set
ZERO_REGISTRY_PUBLIC_KEY_FILEto a PEM file of public keys; with a key configured, unsigned or badly signed entries are refused, and without one nothing changes. - Add support for an external identity provider to handle team sign-in; the built-in OIDC sign-in is used when none is installed.
- Add a documentation page on Open source and Enterprise, and publish the container image under
ghcr.io/0sec-labs/0securityas well as the old name.
Changed
- Change the sidebar to New mission, Findings, Automations and Explore, with Lessons and Computers (when a remote computer is connected) under More.
- Move Skills and Plugins under Explore, move Computers into Settings, and rename Memory to Lessons.
- Move model connections to a Models page with a Providers list at
/models;/connectionsnow redirects there. - Use one vocabulary in the console (Project, Computer, Team and Environment) and rename Scope and autonomy to Permissions.
- Regroup Settings into Preferences, Agent and Activity, and keep privacy and support in General.
- Make Automations use the current project automatically, show This project instead of a project picker, and offer a project folder list from your recent projects when you choose Change.
- Route team authorization through named permissions that a replaceable policy can override; default behaviour is unchanged.
- Run CI checks as parallel jobs behind one required
verifycheck. - Run the native-binary release dry run every night and on changes that affect the binary, using one shared smoke script.
- Generate GitHub Release notes from this changelog, and fail a release that has no changelog section.
Removed
- Remove unused code, duplicate tests and stale assets, including the old workflows demo video and an unused theme preview.
Fixed
- Fix switching project folders so the folder picker stays in sync,
~expands to your home folder, errors are in plain words, and new missions start in the last project folder you used.
Security
- Refuse unsigned or badly signed plugin registry entries when an operator has configured a registry public key.
Install
curl -fsSL https://raw.githubusercontent.com/0sec-labs/0security/main/install.sh | bash
0 --versionWindows (experimental): download 0-windows-x64.exe below. Docker: docker run --rm ghcr.io/0sec-labs/0security:latest --help.
Verify downloads against the attached checksums.txt (SHA-256), for example sha256sum -c checksums.txt --ignore-missing.
Full changelog: https://github.com/0sec-labs/0security/blob/v0.30.0/docs/CHANGELOG.md
0 v0.29.0
0 v0.29.0 (2026-11-14)
Highlights
- New Capy-style console UI with a neutral default accent (orange is optional), Themes, and a slimmer Settings.
- Computers: connect and manage the machines missions run on.
- Skills: Installed and Explore tabs plus a GitHub skills directory with previewed, pinned installs.
- Sentinels: ten presets (Fix Engineer, Secrets Hunter, Release Security Gate and more).
- Recommended models, API key validation (including OpenCode Zen), and visible retry feedback when a provider rate-limits.
- Finding review: Accept records an accepted risk that leaves the review queue but stays visible as Accepted; Dismiss closes a false positive; Undo restores the prior state.
- Fixes for deleting and retrying missions, failed-turn rows, duplicate creates and abandoned drafts.
Install
curl -fsSL https://raw.githubusercontent.com/0sec-labs/0security/main/install.sh | bash
0 --versionWindows (experimental): download 0-windows-x64.exe below. Docker: docker run --rm ghcr.io/0sec-labs/0security:latest --help.
Verify downloads against the attached checksums.txt (SHA-256), for example sha256sum -c checksums.txt --ignore-missing.
Full changelog: https://github.com/0sec-labs/0security/blob/v0.29.0/docs/CHANGELOG.md