Define what you want. Let Claude Code, Codex, opencode, Gemini CLI, GitHub Copilot CLI, and Cursor build it — safely.
Watchfire turns clear specs into scoped tasks, then hands each one to an agent in its own git worktree and sandbox. Better context in, better code out — with clean transcripts and full control at every step.
Releases, deep dives, and the occasional behind-the-scenes note. No spam.
Seven modes that cover interactive chat, single tasks, spec upkeep, and fully hands-off Wildfire runs.
Autonomous loop — execute ready tasks, refine drafts, generate new tasks, repeat.
Interactive session with the coding agent — ask questions, iterate, and pair-program.
Execute a single task from the task list in its own worktree.
Run every ready task sequentially, one worktree per task.
Auto-generate a project definition by scanning your codebase.
Break down the project definition into discrete, ready-to-run tasks.
Fold completed tasks back into the definition — tasks are temporary, the definition is durable.
Watchfire is backend-agnostic. Plug in your favourite CLI coding agent — or mix and match across tasks.
Watchfire turns agents you have to babysit into a fleet you can supervise. The dashboard, notifications, insights, and integrations work together to keep you in the loop without staying on the keyboard.
Aggregate status bar, filter chips, layout toggle, live PTY previews. See working / needs attention / idle / done at a glance, then drill in.
Learn moreTyped notification bus with OS notifications, bundled sounds, dynamic tray menu, and a weekly digest. Headless? They land in a JSONL fallback log.
Learn morePer-task token, cost, and duration capture. Per-project Insights, cross-project rollup, CSV / Markdown export, plus an inline diff viewer pre- or post-merge.
Learn moreOutbound webhooks, Slack, Discord, Telegram, GitHub auto-PR, plus an inbound HTTP server routing slash commands to a transport-agnostic command layer.
Learn morePair your own bot with the daemon and run the fleet from any phone. Just type in the chat to talk to an agent — one starts if nothing is running — and watch mode streams the replies back.
/wildfireStart the autonomous loop from the bus stop and get a milestone feed: generated, implementing, merged./status allOne line per project — what's working, what's idle, what needs you./loginClaude's OAuth token expired? Tap the link, send the code back. No trip to the keyboard.Outbound long-poll only — nothing listens on your machine. Pairing is a one-time code; unpaired chats get silence.
Set up the bridge in five minuteswatchfire_bot
typing…
Isolation, sandboxing, and orchestration — the plumbing that makes multi-agent safe.
Ship work without babysitting sessions. Write tasks in YAML and the daemon handles the rest — picking up ready work, stopping sessions on completion, and chaining to the next task. Every run lands in the same clean transcript, no matter which backend produced it.
Every task runs in its own git worktree on a dedicated `watchfire/<n>` branch — never the working tree you're in. Tasks run one at a time per project (the agent finishes one before starting the next), but each one is isolated from your branch and from any work in flight. Finished tasks auto-merge back to your default branch.
Your credentials and dotfiles stay off-limits, even if an agent goes off-script. Every session runs inside a platform-native sandbox — Seatbelt on macOS, Landlock or Bubblewrap on Linux — blocking ~/.ssh, credential stores, and git hooks by default. Restart protection stops runaway loops before they burn budget.
Drive every repo on your machine from one place. A single daemon runs multiple projects concurrently — each with its own task queue and one active agent at a time — so spreading work across repos is how you get real parallelism. The system tray keeps them all one click away.
Same agents you’d run yourself — wrapped in the safety rails you’d otherwise build by hand.
Edits land directly on your working tree; one bad refactor and you're fighting git stash.
Each task runs on its own git worktree and branch. Failed runs leave your main branch untouched.
The agent inherits your shell's PATH, SSH keys, and cloud creds. One prompt-injected command, one bad call.
Sandboxed by Seatbelt on macOS and Landlock on Linux; ~/.ssh, credential stores, and .git/hooks blocked by default.
One repo, one agent, one terminal. Three tasks waiting on the previous to merge.
One agent per project, but every project on your machine runs concurrently — open several repos and they all make progress at once.
Scrollback in your terminal is the only record. Hope you didn't clear it.
Clean transcripts captured per task; inline diff viewer on the worktree branch; insights track duration, tokens, and cost.
You manage what runs when, by hand.
Wildfire mode runs ready tasks, refines drafts, and generates new ones toward your project definition.
Concrete moments where Watchfire earns its keep — not features, but the work itself.
Each project runs one agent at a time, but every project on your machine runs concurrently. Queue ready tasks in three repos and watch them drain in parallel — each on its own worktree, each auto-merging back when it's done.
Read moreHand a long refactor to a task and keep moving on the next thing in chat. The session runs in the background; the transcript and the diff are waiting when you come back, ready to review.
Read moreThe Beacon dashboard rolls every project into one window — what's running, what needs attention, what's done. Slack, Discord, Telegram, and OS notifications fan out the rest — pair the Telegram bridge and you can watch and answer your agents from your phone.
Read moreWildfire mode picks up ready work, refines the drafts, and writes new tasks toward your project definition until the goal is met. Walk away with a backlog; come back to a branch full of completed work.
Read moreInstall in seconds. Define tasks. Let agents ship code for you.
Download for macOSAvailable for macOS, Linux, and Windows
# Install via Homebrew (macOS)brew tap watchfire-io/tapbrew install --cask watchfire-io/tap/watchfire# Set up your project and gowatchfire initwatchfire task add "Build the login page"watchfire start --allIncludes GUI, CLI, and daemon. Also available via Homebrew.
Watchfire.app puts your repos, tasks, branches, and live agent terminals in a multi-window desktop client — one window per project, one mission-control dashboard for the fleet — built on the same daemon as the CLI.
Each project window is chat-primary: the agent terminal is the wide left pane, with everything else in a tabbed reference panel on the right — so the live work surface leads and the paperwork stays one tab away.
Home window
Mission control for the fleet — the dashboard rolls up every project, flags what needs attention, and opens each one in its own window. An always-on-top mini-monitor keeps it glanceable.
Chat pane
The agent terminal, front and center — pick a backend, start any mode including Wildfire, and watch it work. A built-in terminal stays docked at the footer.
Reference panel
Tasks, Definition, Insights, Secrets, Trash, and Settings in tabs on the right. Hide it when you want the chat full-width, pop it back when you need the paperwork.
A daemon orchestrates in the background while the CLI and GUI give you full control.
Project-scoped commands plus a rich interactive TUI with split-pane task navigation and live output.
Electron multi-window client with a native feel — one window per project, a mission-control dashboard, and embedded live terminals.
Background service handling orchestration, PTY management, git workflows, and the gRPC API.
The honest version — what Watchfire does, what it doesn’t, and how it stays out of your way.
sandbox-exec) on macOS, Landlock (kernel 5.13+) with a Bubblewrap fallback on Linux — with ~/.ssh, credential stores, and .git/hooks blocked by default. Restart protection caps runaway loops before they burn budget. Full sandboxing details.watchfire) and the interactive TUI cover everything the GUI does — they all talk to the same daemon over gRPC. The Electron GUI is just another client; pick whichever fits your workflow, or use both side by side.watchfire/<n>) and post-merge (reconstructed from the merge commit), and a per-task Insights record captures duration, tokens, and cost. GUI walkthrough · Insights & metrics.Notes, releases, and deep dives from the team building Watchfire.
7 min
Torch pairs a Telegram bot with the daemon so you can supervise every project from your phone — and not just with commands: just type in a paired chat and you're talking to a chat agent, with the replies streaming back. It works with no public endpoint, no tunnel, and no port forward, because the daemon dials out. Around the tentpole: a definition-retrofit mode, batch quick-add, and a burn-down of long-standing friction.
3 min
Two GUI issues had been sitting open: every Nerd Font icon in the embedded terminals rendered as a missing-glyph box, and the notification bell was a read-only list with a count that only ever went up. 9.4 closes both. The font fix is a small lesson in how Chromium resolves glyphs; the notification fix is a small lesson in never shipping a badge without a way to clear it.
4 min
Every non-Claude agent session materializes a scratch home directory under ~/.watchfire — and until now, nothing ever deleted it. One directory per project × mode × task, forever, surviving even project deletion. 9.3 fixes the leak twice: sessions clean up after themselves, and deleting a project sweeps whatever its sessions left behind. The interesting part is deciding what not to delete.
Install in seconds. Define tasks. Let agents ship code for you.
Download for macOSAvailable for macOS, Linux, and Windows
# Install via Homebrew (macOS)brew tap watchfire-io/tapbrew install --cask watchfire-io/tap/watchfire# Set up your project and gowatchfire initwatchfire task add "Build the login page"watchfire start --allIncludes GUI, CLI, and daemon. Also available via Homebrew.
brew tap watchfire-io/tap && brew install --cask watchfire-io/tap/watchfire