Foxpit 0.1.12

Foxpit Pro is here: upgrade from the app or from your account on foxpit.dev ($5 for the first month, then $12/month). The Projects screen shows your plan and what your agents are doing.

Added

• Your account on the web: foxpit.dev/account.html shows your plan, trial days left and devices, and lets you upgrade or manage your subscription. Sign in with the same Google, GitHub or email code you use in the app
• Foxpit Pro checkout: "Get Foxpit Pro" and Account → "Upgrade to Pro" open the payment page in your browser ($5 for the first month, then $12/month); Foxpit unlocks Pro on its own once the payment goes through. Pro users get "Manage subscription"
• Projects screen: your account, plan and trial days left sit in the top right (opens Account). When the trial is about to end or has ended, a banner offers Upgrade to Pro
• Projects screen: rename a project from its ⋯ menu; projects still called "New project" show their folder name
• Projects screen: cards show "2 working" or "Needs you" while agents run there

Changed

• Projects screen: Live sessions lists only agents that are working, waiting for you or failing (idle tabs no longer fill the list); cards hide zero counts and say "No activity yet"

Fixed

• The sign-in code error now says 8 digits
• Account window: the "Account" label no longer stretches across the window

Foxpit 0.1.11

Beta: one team setup everywhere, full model names, and Foxpit engine works with any AI provider (API key, subscription or your own server). Found a problem? Settings → Report a bug.

Added

• Team setup: the first-run wizard, "+ New project" and Manage Team share one team table. Tick a mascot, give it a specialty and pick which agent and model it runs on; Foxpit looks at the project folder and suggests a team, but only applies it when you press "Use suggestion"
• Manage Team → "Suggest a team" opens the same table for the current project, with your current team (including helpers already in the office) selected
• "+ New project" asks who leads (Fox's agent and model) and has an "Open their terminals now" checkbox instead of a helper counter
• Agents that are installed but not signed in on this computer get a "Not signed in" badge; they can still be picked
• When a model is left empty, Foxpit shows the model the agent will really use, e.g. "opus (default)", read from Settings or the agent's own settings file
• Model names are written out in full everywhere you pick or see a model: "Claude Opus 5.5" instead of "opus", "OpenCode Go · DeepSeek V4.1 Flash" instead of "deepseek-v4.1-flash". Claude Fable 5.1 is in the Claude list too
• Foxpit engine works with any AI provider: pick it as an agent and Foxpit shows which providers are connected. "Connect a provider" opens the engine's provider settings right there: paste an API key (OpenRouter, OpenAI, DeepSeek, Anthropic…), sign in with a subscription, or add your own server (Ollama, LM Studio, any OpenAI-compatible address). Keys stay in the engine on this computer. If the engine isn't set up yet, it can be set up from the same place

Changed

• A new project always asks where it lives: "Use an existing folder" or "Start a new empty folder", which shows the exact path it will create (e.g. C:\Users\you\Foxpit\My app). Foxpit no longer creates a folder silently when you skip the choice; "Start project" and the wizard's Next point you to the folder choice instead

Fixed

• After an automatic update restarted Foxpit, the next automatic update was switched off for that session (the restarted app thought it was a test copy) and it could not move to a free port if its usual one was taken
• A teammate whose model isn't in the list (typed by hand in Manage Team) showed "Same as Fox" in the team table while a different model was saved
• "Suggest a team" could save an office helper's live agent together with a model meant for another agent
• "Same as Fox" now names the model the helper really gets (Settings → Default model per agent comes first)
• OpenCode showed "Not signed in" when its first login file was empty or half written; Claude Code's and Codex's default model is read even from files saved with a BOM or an escaped quote
• Models on your own server (Ollama, LM Studio, custom endpoints) never showed up for Foxpit engine teammates, and handing them work could not pick that server
• OpenCode 2 cards showed the session's running token total as context (e.g. 92% while OpenCode itself said 7%); context now comes from the latest request

Foxpit 0.1.10

Beta: OpenCode 2 works again, images paste into agents, and Foxpit keeps itself and your agent tools up to date. If you're on 0.1.9 or older, download this version once by hand: automatic updates were switched off in those versions. Found a problem? Settings → Report a bug.

Added

• Chat tabs (OpenCode Chat, Gemini Chat, Hermes Chat) send pasted or dropped images as real image content (base64 image blocks) instead of a plain file path line; other files still go as a path and the agent reads them itself

Fixed

• Foxpit and agent CLI auto-updates never ran: a startup setting made every copy look like a test instance. Both now run as intended.
• Pasting a screenshot into an agent terminal (Claude Code, Codex, OpenCode) did nothing. Ctrl+V now attaches the image, and so does dropping a file onto the terminal. Files copied in Explorer paste as attachments too
• The chat view takes images as well: paste or drop one and it shows above the message box; it's sent with your message
• The chat input falls back to the system clipboard when a paste event carries no file list, so a raw screenshot attaches on every path
• OpenCode 2 showed "Plugin failed" and Foxpit saw nothing from OpenCode tabs. Foxpit now installs a plugin built for OpenCode 2 (OpenCode 1 keeps the old one), so live output, tool activity, token use and subagents show up again
• Opening an OpenCode 2 tab with a chosen model failed with "Unrecognized flag: -m"; the model is now passed the way OpenCode 2 expects, and each tab runs its own OpenCode server so its activity reaches the right desk
• Handing work to an OpenCode agent failed on OpenCode 2 ("Unrecognized flag: --dir")
• A chat agent that can't take images now gets the image as a file path instead of rejecting the message
• Attachment chips showed the whole Windows path instead of the file name
• Agent tools never updated on Windows while one was open (Claude Code is almost always open), so new models like Claude Haiku 5.5 never arrived. Foxpit now updates Claude Code, Codex, OpenCode, Gemini CLI and Antigravity even while they run: the files in use are moved aside, the new version is installed, open tabs keep working on the old copy, and the old copy is removed later. If an update fails, the old version is put back
• Each tool is updated the way it was installed (npm, its own installer, or scoop); tools installed with another package manager are left alone and the reason is shown in Settings
• Updates no longer leave a ~230 MB leftover folder in the npm folder each time; leftovers from earlier updates are cleaned up
• No connection, a timeout or files in use no longer wait a day: Foxpit tries again on the next check. Two Foxpit windows never update at the same time
• Foxpit checks for tool updates every few hours while it stays open, not only at launch, and tells you when a tool was updated
• Opening a tab for a tool that is being updated waits for the update to finish instead of failing

Foxpit 0.1.9

Beta: new work desks that show how each agent is doing. Found a problem? Settings → Report a bug.

Added

• Each work desk has a light strip that shows the state of the agent sitting at it: the agent's own colour, breathing while it works; yellow while it waits for your approval; red on an error; green when it's done; dim when it's idle. While an agent is busy, its desk also glows on the floor, so you can spot it from across the office
• With Reduce motion on, desk lights keep their colour but stop pulsing

Changed

• New work desks with a walnut top and a metal frame replace the old boxy ones. Desks you have moved keep their place, monitors and chairs

Foxpit 0.1.8

Beta: calmer updates with a "What's new" view, and Mac updates that install themselves. Found a problem? Settings → Report a bug.

Added

• Updates no longer interrupt you. A small chip in the top bar shows the download ("Updating 45%"), then "Update ready". Click it to see what's new, then restart now or later; later installs it when you quit
• Installing an update shows a short "Installing update…" screen and Foxpit opens again by itself, with no installer window. After the update, a note shows the new version and what changed
• Mac: updates now download and install themselves. Foxpit replaces itself when you restart or quit, then opens the new version. If Foxpit runs from the disk image or a folder it can't write to, you still get the "Update available" notice

Foxpit 0.1.7

Beta: Mac fixes, a live office and lower memory use. Found a problem? Settings → Report a bug.

Added

• The back wall of the office is now a team board: who is working and on what, today's deliveries, a live activity feed and token use. Click it to zoom in
• Desk monitors show what each teammate is working on, live
• Office off switch: turn the 3D office off to free memory while your agents keep working
• Agent tools (Claude Code, Codex, OpenCode) and the Foxpit engine are kept up to date in the background, once a day; Settings shows each tool's version and result, and you can turn it off
• Sub-agents get their own context and token count instead of adding to their parent's
• Working mascots mumble now and then, with a new, softer voice

Changed

• Foxpit uses about 170 MB less memory at launch

Fixed

• Mac: the window buttons sat on top of the tab bar, top bar and Projects title
• Mac: sounds stayed silent after the app woke up
• The Chat/Terminal button covered the terminal's first line, and "Start an agent" could stay over a live terminal after a reload
• In a narrow office panel the room was cut off
• The usage pill shows "–" instead of a bare icon when there is no data yet
• Memory grew each time the office was turned off and on again
• Room edits were lost after the office was turned off and on again
• A tool update that stopped halfway could leave Claude Code shown as "Not installed"; Foxpit now repairs it and skips tools that are in use

Foxpit 0.1.6

Beta: fixes from a fresh-install test. Found a problem? Settings → Report a bug.

Changed

• Setup no longer preselects teammates. When you add one (or click "Use suggestion"), you must choose the agent and model it runs on before the office opens; the list shows both, e.g. "Claude Code · sonnet"

Fixed

• Clicking a mascot in full screen opened its card away from the mascot, over the chat panel
• Teammates' desk screens showed "CLI exposure failed": the engine no longer edits PATH or its launchers when Foxpit starts it, and the line is hidden for other engines
• Foxpit Chat: a sub-agent's tokens could be counted on the chat's mascot instead of the sub-agent's own
• Foxpit Chat: its font was blocked, leaving a security error in the log

Foxpit 0.1.5

Beta: Foxpit now looks at your project and suggests a team, and shows you your first handoff. Found a problem? Settings → Report a bug.

Added

• Team suggestion: when you pick a folder, Foxpit scans it in about a second and suggests which mascots and roles fit, with a reason for each. Nothing is applied until you click "Use suggestion", and you choose the agent and model for every teammate
• Your first task: after setup, a short guide offers tasks that fit your project, sends your pick to Fox, then shows the handoff to a teammate and the delivery
• Delivery cards: when a teammate finishes, a card lists changed files, test results and cost (or "Cost unknown"); a Deliveries tab keeps today's history
• Usage panel: real plan quota, today's spend, a limit card and automatic resume
• Livelier chat: new words fade in, the running step glows, finished steps get a tick and the plan shows n/m progress

Changed

• Calmer office: status pills hide after work is done and desk screens sleep while an agent is idle; repeated output lines are merged

Fixed

• After finishing setup with a chosen folder, an empty Projects screen could stay on top of the office
• A label ("Claude") could be saved as a teammate's model name
• A status pill could overlap another agent's open desk screen

Foxpit 0.1.4

Beta: accounts and the 3-day free trial. Found a problem? Settings → Report a bug.

Added

• Sign in with GitHub, Google or an email code; a 3-day free trial starts on first sign-in, no card needed
• Trial badge in the title bar and a lock screen when the trial ends
• Projects: a Remove button for damaged saved records

Fixed

• Sign-in could fail with "Sign-in failed" right after GitHub returned to the app
• Restored terminal tabs no longer show as "working" just because they printed their startup output
• Sign-in errors are now shown on the return page and written to the log

Foxpit 0.1.3

Beta: the first release with Foxpit Chat and its built-in engine. Found a problem? Settings → Report a bug.

Added

• Foxpit Chat: the built-in engine ships with the app; first-time users without any CLI can pick "Foxpit engine" in setup
• Desk screens show each agent's real steps (commands, output, edited files); the lead's screen is open by default
• Who-waits-on-whom labels, agent to-do progress on cards, context-full warning, live sessions board on the Projects screen
• Replay highlights, recap as Markdown and video export; a sandboxed demo office from the welcome screen
• Agent voices (on by default): mascots mumble when they hand work to each other
• Mascot-free view: Settings → Office → View turns the 3D office off for a plain chat and terminal app
• Foxpit engine members get a real model list from the engine

Fixed

• No sound at all until a key was pressed in the office
• Foxpit Chat tokens, context % and Hermes sub-agent spend now count in the office
• A fresh engine no longer fails its first reply when only a non-Anthropic key is present
• Settings files (including Claude/Codex settings) are written atomically, so a crash cannot leave them half-written
• Finished sub-agent output can be reopened from its mascot
• The 3D engine pauses while the Projects screen covers the office or mascots are off (much lower CPU)

Windows

• SmartScreen may warn because the app is not code-signed yet: More info → Run anyway