v2.0.14 · MIT licensed · No telemetry

Every client. One agent workforce.

CodeFire is the desktop command center for agencies and teams running AI coding agents. Team chat with your agents, orchestration across every project, client-ready reporting, and shared memory — one place for Claude Code, Codex, Kimi Code, and the CodeFire Agent to work like a team.

codefire://install-mcp?client=claude  ·  one-line MCP setup
CodeFire projects dashboard — 105 projects grouped by client with live agents, open tasks, and weekly progress on every card
Scroll
Workbench for —
CLAUDE CODE
CODEX
KIMI CODE
CODEFIRE AGENT
+ ANY MCP-AWARE AGENT
CLAUDE CODE
CODEX
KIMI CODE
CODEFIRE AGENT
+ ANY MCP-AWARE AGENT
Chapter I · The workbench

Stop juggling clients.
Start shipping with
an agent workforce.

A chat-first shell built for client work. Your team's conversation with its agents stays front and center while a rail of dockable tools — Tasks, Workflows, Browser, Wiki, Notes, Git, Sessions, Files, Terminal, Activity, Reports — opens beside it or tears off into floating windows. Four CLIs, every client project, shared memory.

Team chat with agents
Channels where you @mention agents and work routes itself — runs, approvals, and threads inline.
Agent teams & orchestration
Custom agents grouped under one @handle — relay or parallel — plus pipelines and cross-CLI handoffs.
Multi-client command center
Group projects by client, watch every live agent, and keep every board moving from one dashboard.
Client reporting & memory
Branded PDF reports and invoices from completed work — backed by wiki memory agents keep fresh.
Chapter II · Team Chat

Mention an agent.
Work happens.

Every project gets Slack-style channels where humans and agents share the room. @mention an agent — or a whole agent team — and CodeFire routes the work, tracks it as a task, and posts the live run right in the conversation.

CodeFire Team Chat — a #team-chat channel with completed Plan and Implement agent runs, an agent team finishing 14 legs of work, and a threaded session summary
@
Route by mention
@claude, @codex, @kimi — or any custom agent handle. Explicit mentions launch runs; prose never does.
Agent teams
Group agents under one handle. Run them in parallel, or relay in order — plan, implement, review.
Approvals inline
Permission prompts surface in the channel with live run cards — elapsed time, tokens, stop and retry.
Rosters that sync
Publish an agent team once and every teammate imports it automatically — same lineup, same order.
Chapter III · The three engines

One app. Three
engines.

01 — Workflows & Pipelines

Pipelines,
not prompt chains.

Save a procedure once, run it any time — "Onboard a new client", "Ship a release", "Security review pass" — and agents execute every step in sequence. Or compose multi-agent pipelines: plan with one agent, implement with another, verify with a third, each step routed to the CLI best suited for it.

Workflow templates Per-step agent routing Scoped auto-approve Worktree isolation
02 — Agent Handoffs

Claude starts.
Codex finishes.

Hand a running conversation from one CLI to another — or to a human teammate — with context, model selection, and task state carried over. A live agent census shows every running agent across every client project, so nothing works in the dark.

Live agent census from:claude → to:codex Approve / re-target Shared task notes
03 — Context Engine

Surgical retrieval,
not file dumps.

Hybrid vector + keyword search across functions, classes, blocks, docs, and git commits — exposed to your agent as context_search. Stale results never block; the index re-verifies in the background after sessions and on a scheduler.

Vector + FTS Git history aware Wiki + code ranked Background refresh
The task board

Every client,
on one board.

Kanban that fills itself. Work you hand off in Team Chat is tracked automatically — every agent run attaches to a task, so client requests never evaporate into a chat log. Pull in work from email and mobile capture, then watch agents move cards to Done.

  • Tasks auto-created and tracked from Team Chat
  • Global Kanban across every tracked project
  • Staging inbox from Gmail and mobile capture
  • Stale-task cleanup with a full audit trail
CodeFire Tasks board — 52 tasks across To-do, In Progress, and Done, with agents running and tasks auto-tracked from Team Chat
CodeFire agent chat — a live session streaming edits, with model picker and per-session permissions in the composer
Agent chat

Pick an agent.
Start working.

Launch a real Claude Code, Codex, Kimi Code, or CodeFire Agent session against the open project — supervised, streamed, and scoped to the folder you chose. Set permissions per session, define custom agents with their own instructions and models, and hand work off without leaving the app.

Claude Code Codex Kimi Code CodeFire Agent
  • Per-session permission mode: Ask · Plan · Auto edits · Full access
  • External context folders you can attach on the fly
  • Output streams back into chat — replayable
Wiki memory

The brief
that survives
every session.

Architecture decisions, launch runbooks, client gotchas, glossaries — written once and read by every future session, on every teammate's machine. CodeFire seeds the structure from your repo, agents author through MCP, and freshness metadata flags pages that drift. New developer on the account? The brief is already written.

  • One-click init: Overview, Stack, Architecture, Schema, Runbook
  • [[Wiki Links]] + backlinks for connected knowledge
  • Freshness statuses — current, draft, stale, superseded — plus one-click "mark verified"
  • Local-first SQLite. Optional team cloud sync.
CodeFire Wiki — a client project's launch-readiness page, verified 56 minutes ago, with related files and a sidebar of current pages
CodeFire project terminal running the real Claude Code CLI inside the workbench
Watch agents work

Real CLIs.
Real output.

CodeFire spawns the actual CLI binaries you'd run in your terminal — no wrappers, no scraping. Tool calls, shell commands, and assistant turns stream into the chat panel as they happen, ready for you to approve, pause, or pivot.

  • Approval prompts surfaced inline — not buried in a TTY
  • Per-project defaults for agent, permissions, and context
  • Full session transcripts saved and searchable
MCP server

126 tools.
One protocol.

Every CodeFire surface — wiki, tasks, notes, sessions, context search, browser, terminal, images, workflows, git, agent handoffs — is exposed to your CLI through a single MCP server. Agents don't just read your codebase; they update the project's shared brain.

Wiki Tasks Notes Context search Browser Terminal Handoffs Sessions Images Workflows Git Reporting
A task created through the CodeFire MCP server — status, priority, labels, and auto-run dispatch
CodeFire portfolio and team reports — a branded Workspace Work Distribution PDF with work by client and by project, scheduled to generate monthly
Client reporting

Reports your clients
actually read.

Branded, letterhead PDF reports built from the work itself. Completed tasks become an invoice ledger with AI hour estimates, rates, tax, and Net-30 terms — or roll the whole portfolio up by client, by developer, or by team, on a schedule.

  • Branded letterhead orgs — logo, address, accent color
  • Invoices from completed tasks with AI hour estimates
  • Portfolio roll-ups: by client · by developer · by team
  • Scheduled weekly, monthly, or yearly PDF generation
Connect Google

Your clients' numbers,
on tap.

Connect Google once and assign GA4 properties and Search Console sites to each client project. Analytics report recipes turn them into executive summaries, KPI scorecards, and trend tables — and your agents can query the same data through MCP. Credentials stay encrypted on your machine, never exposed to agents.

  • GA4 traffic, engagement, and top pages per project
  • Search Console queries, clicks, and positions
  • PageSpeed audits with ranked opportunities
  • Gmail to tasks · Calendar agenda with Meet links
CodeFire reporting connections — Connect Google, assign GA4 and Search Console properties to a project, and build analytics report recipes
CodeFire Capture · mobile companion

Your agents,
in your pocket.

CodeFire Capture mirrors your agent chat to your phone. Watch runs stream live, capture tasks and ideas from anywhere, and steer an in-flight agent from the couch — the desktop stays the source of truth.

  • Live mirror of agent chat, run status, and approvals
  • Capture tasks on the go — they land in your staging inbox
  • Remote control is opt-in, per device, and revocable
Chapter IV · And the rest of your dev stack

Browser. Terminal.
Git. Images.

CodeFire isn't only a memory layer — it's a desktop dev environment for the agentic era. The things you'd normally tab between, in one window your agents can also reach.

CodeFire Images drawer — AI image generation with aspect-ratio presets and media history
CodeFire Transcripts — record a client call or voice note and CodeFire transcribes it and pulls action items onto the task board
Persistent browser
In-app browser your agents automate — tabs, navigation, screenshots, network capture.
Tabbed terminal
Real pty sessions alongside your project views, with show/hide toggle and CLI orchestration.
Git intelligence
Status, diffs, staged changes, and branch context — exposed to agents and to you.
AI image gen
Generate and edit images for design comps, social, hero shots — saved to project media.
🎙
Talk into tasks
Record a client call or voice note — CodeFire transcribes it and stages the action items as tasks.
Notes & briefings
Pin architecture decisions and gotchas. Get AI-generated daily project briefings.
📥
Gmail + capture
Whitelist email senders into tasks. Send tasks from the CodeFire Capture mobile app.
Session history
Token usage, cost, tool calls, file changes — every run, observable.
126
MCP tools exposed
4
Agents, one workbench
0
Telemetry collected
MIT
Open source license
Chapter V · Bring your own agent

Four first-class agents.
Any MCP client.

Claude Code, Codex, Kimi Code, and the built-in CodeFire Agent run as first-class in-app agent chats with live handoffs between them. Any other MCP-aware CLI still gets full access to the 126-tool MCP server.

Any other MCP-aware agent connects through the same 126-tool MCP server — setup guide.

Chapter VI · Get CodeFire

Download it.
Light the fire.

Free and MIT licensed. Signed, notarized builds straight from GitHub Releases — updates are checked there too, never silently installed.

download codefire Step 1

Grab the latest release for your platform. macOS builds are signed and notarized — no Gatekeeper hoops. Apple Silicon only; Linux ships x86_64 and arm64.

Full install walkthrough in the getting started guide.

connect your agent Step 2

One-click MCP install for any supported CLI — these deep links open CodeFire and wire up your config automatically.

Questions

Frequently asked.

Is CodeFire free?

Yes. CodeFire is MIT licensed and free to download. There is no telemetry — your projects, tasks, and wiki live in a local SQLite database on your machine.

Which AI coding agents does CodeFire support?

Claude Code, Codex, Kimi Code, and the built-in CodeFire Agent run as first-class in-app agent chats with live handoffs between them. Any other MCP-aware CLI can connect through the bundled 126-tool MCP server.

Can CodeFire manage multiple clients and projects?

Yes. Projects group by client, the dashboard shows every board, open task, and live agent at once, and portfolio reports roll up completed work by client, by developer, or by team — exportable as branded PDFs.

How does Team Chat with agents work?

Every project gets Slack-style channels where humans and agents share the room. Mention an agent — or a saved agent team — and CodeFire routes the work, tracks it as a task, and posts the live run inline with approvals and threads. Shared agent teams sync to every teammate.

Does CodeFire run on Intel Macs?

The macOS build is Apple Silicon (arm64) only. On Linux, both x86_64 and arm64 are supported via AppImage and .deb packages.

When is the Windows version coming?

A Windows build is in the works. and we will email you the moment it ships.

Does CodeFire collect telemetry?

No. CodeFire collects zero telemetry. Everything is local-first; team cloud sync is optional and off by default.

What is the CodeFire MCP server?

CodeFire bundles an MCP server that exposes 126 tools — context search, wiki, tasks, notes, sessions, browser automation, terminal, images, workflows, git, and agent handoffs — to any MCP-aware coding agent.

Built by an agency that runs agents on every client and got tired of re-briefing all of them.