For agencies & teams · Multi-client collaboration

Run every client
with an agent team.

Give every client project a support crew of agents your whole team can mention, orchestrate, and audit. Shared channels, shared agent rosters, shared memory — and reporting your clients can actually see.

Download CodeFire Getting Started

Chapter I · The cost of chaos

Multiplied across
every client.

01 — Time

Context dies between sessions

Every agent session starts by re-discovering what the last one already knew — multiplied by every developer, on every client account.

02 — Knowledge

Knowledge stays siloed

Dev A fixes a tricky deployment issue on one client. Dev B hits the same wall next week on another. Their agents have no way to share what was learned.

03 — Visibility

No visibility into AI work

Leads can't see what agents are doing across accounts — what's in progress, what was decided, what it cost, or what's blocking a launch.

Chapter II · Team Chat

One room.
Humans and agents.

Every client project gets Slack-style channels — plus a cross-project agency channel — where teammates and agents share the room. @mention an agent or a saved agent team, and the work routes, tracks itself as a task, and posts the run inline.

→ Invite teammates into channels with owner/member roles
→ Agent teams run in parallel or relay — plan, implement, review
→ Approvals, threads, and live run cards in the conversation
→ Chimes when a team finishes or a teammate messages

CodeFire Team Chat — humans and agents in a #team-chat channel with completed agent-team runs and a threaded session summary

CodeFire Skills — a shared skill library available to Claude Code, Codex, and the CodeFire Agent

Chapter III · Shared agent teams

Build the crew once.
Everyone gets it.

Define custom agents — provider, model, instructions, permission mode — and group them into named teams under one @handle. Publish a team to your workspace and every teammate imports it automatically: same lineup, same relay order, same behavior.

→ Custom agents: objective, instructions, model, permissions
→ Shared rosters sync to teammates via team cloud
→ Skills shared across Claude Code, Codex & CodeFire Agent
→ Global or per-project scope — one client or all of them

Chapter IV · Client groups

A hundred projects.
Ten clients. One view.

Group projects by client with colors and billing rates, then run the whole book of business from the dashboard — progress bars, open tasks, live agents, and this-week momentum on every card.

→ Client groups with per-group colors and rates
→ Filter the dashboard by group, name, or task load
→ Stale-task cleanup with an audit trail per client
→ A staging inbox that catches tasks from email and mobile

CodeFire Groups — ten client groups with colors, project counts, and per-group billing controls

CodeFire Files drawer previewing a client-facing go-live readiness report written by an agent team

Chapter V · Client deliverables

Agents that write
for the client.

Launch audits, go-live readiness briefs, status reports — agent teams produce client-ready documents right in the project, previewable in the Files drawer. Pair them with branded PDF reports and invoices built from completed work, on a schedule.

→ Portfolio reports: work by client, developer, or team
→ Invoices from completed tasks with AI hour estimates
→ GA4 + Search Console analytics recipes per client
→ Scheduled monthly PDFs — generated while you sleep

Chapter VI · Visibility

Every run,
accounted for.

Session history captures every agent run — model, duration, token breakdown, files changed, and cost — with AI summaries you can share to the team. The activity feed shows who did what across every account, in real time.

→ Token and cost tracking per session, per project
→ AI summaries · extract tasks · share a session to the team
→ Team activity feed across tasks, notes, and runs
→ Role-based access: lead, contribute, or view

CodeFire Sessions — searchable run history with token breakdowns, AI summaries, extract-tasks, and share-to-team

Chapter VII · Shared by default

Everything your
team needs.

Shared task board

Tasks, in sync.

Drag-and-drop Kanban visible to the whole team. When one agent creates or completes a task, every team member's agent sees it instantly.

Notes & decisions

Pin once, share forever.

Architecture decisions, gotchas, and bug patterns captured as persistent notes. Pin critical context so every agent reads it at session start.

Wiki memory

The brief that survives.

Per-client wikis with freshness tracking, synced across the team. A new developer on the account starts with the brief already written.

Semantic search

Search by meaning.

Vector + keyword hybrid search across your indexed codebase. Every agent on the team can search by meaning, not just string matching.

Activity feed

Real-time visibility.

Real-time visibility into what's happening across the team — task updates, note creation, session activity, and code changes.

Role-based access

Lead, contribute, view.

Invite team members as leads, contributors, or viewers. Project visibility is scoped — team membership alone never exposes every client.

Chapter VIII · Up & running

Three steps,
minutes to ship.

Download CodeFire

Grab the latest release for macOS or Linux, then create your team in-app and invite developers by email.

Connect your CLI

Each developer adds the CodeFire MCP server to their AI coding CLI — one-click install for Claude Code, Codex, Kimi Code, and the built-in CodeFire Agent.

iii

Assemble the agent teams

Group projects by client, publish your agent rosters, and start mentioning them in Team Chat. Tasks, notes, wikis, and session context sync from the first run.

View the full Getting Started guide →

Run every client with an agent team.

Multiplied across every client.