An open-source AI team that ships real work — with a conversation you can follow.
Website & 58s narrated intro · Quickstart · How it works · What's verified · 日本語
▶ Narrated intro (59s, English) · 日本語版 (58s). The GIF above drives the real UI and runtime with the scripted test provider (no LLM calls, labelled “FAKE PROVIDER” on screen) so it is deterministic and free to reproduce. With a real connection the same screens are fed by live model calls; the run header shows the model the provider actually reported and the measured cost.
You type one request. A Master plans the deliverables, a Researcher, a Builder and a Reviewer actually do the work, and you get the files plus the real bot-to-bot messages, a timeline, and verification bound to each artifact revision.
- No scripted chat. The chat panel is a projection of
message.sentevents — messages that were really delivered to another bot's mailbox. A question wakes the other bot to answer. - Evidence, not vibes. Checks and review verdicts are events bound to an artifact revision hash. The final report is compiled from the event log; a model summary cannot upgrade “started” to “done”.
- Runtime-enforced limits. Tool scope, write scope, budget reservation, approvals and cancellation are enforced in code, not by prompt wording.
- Per-bot configuration. Each bot inherits a default connection and model and can override endpoint, model, effort and system prompt (lockable). Configured vs. provider-reported model are both shown. No silent fallbacks.
- Replay, resume, fork, export. Replay never calls a model. Fork from a checkpoint with a different model for one bot and compare. Export the whole run as JSONL.
No API key needed if you have Claude Code installed and logged in — the default connection runs every agent session through the local claude CLI (its cost/usage and the model it actually used are read from the CLI's JSON result).
git clone https://github.com/FORIFOR/Multibot && cd Multibot
cd backend && uv venv .venv --python 3.12 && uv pip install --python .venv/bin/python -e '.[dev]'
cd ../frontend && pnpm install && pnpm build && cd ../backend
.venv/bin/agentteam probe # real capability check through `claude -p` (a few cents)
.venv/bin/agentteam serve # http://127.0.0.1:8787Prefer the API? Set defaults.connection_id: anthropic in data/agents.yaml and export ANTHROPIC_API_KEY (the config stores only the reference env:ANTHROPIC_API_KEY). Any OpenAI-compatible chat endpoint and local Ollama (driver: ollama, http://localhost:11434/v1) work the same way.
First run: Settings → Probe (or agentteam probe). It confirms tool calling and JSON-schema output on the model you picked, then unlocks Start. Placeholder models, unknown prices and unverified connections refuse to start with a structured reason.
How the CLI path works: Claude Code owns the agent loop for one session; the team's tools (send_message, publish_artifact, run_check, …) are exposed to it as MCP tools through a stdio proxy that forwards every call to this process's ToolGateway, so policy, budget and the event log are identical to the API path. Claude Code's own built-in tools are disabled for these sessions.
CLI:
.venv/bin/agentteam validate # config check
.venv/bin/agentteam probe [connection] --model ID # real capability probe
.venv/bin/agentteam run "Turn this into a launch page and 3 posts" --url https://example.com --budget 1.5| Component | Responsibility |
|---|---|
| Master (planner) | Request → deliverables, recorded assumptions, task DAG (structured output). The runtime validates schema, cycles, owners, tools, write scopes and limits; invalid plans are sent back once, then the run fails honestly. |
| AgentRunner | One bot session = its own conversation state + a tool loop through the gateway. A worker receives its task, input artifact refs and its own inbox — not the whole history. |
| MessageBus | send_message really delivers to the recipient's mailbox and records message.sent. Purposes are limited to request / question / answer / handoff / finding / decision. |
| ArtifactStore | publish_artifact creates an immutable, SHA-256-hashed revision with an atomic write. Reviews and checks bind to a revision. |
| Scheduler | Dependency resolution, concurrency limit, review fail → revise → re-review (bounded), Master exception decisions, checkpoints. |
| PolicyEngine | Budget = spent + reserved for in-flight calls; call/tool/message limits; tool and path scope; cancellation. Unknown cloud prices are never treated as free. |
| Sandbox | macOS sandbox-exec (no network, writes only inside the task workspace). Elsewhere a plain subprocess that says so in every result. |
| Providers | claude_cli (local Claude Code, no key), anthropic_messages (official SDK), openai_compatible_chat, ollama. Per-bot choice; real probe before start. |
Everything is written to an append-only event store (SQLite WAL, per-run seq, UTC timestamps, redaction before persistence). Chat, timeline and report are projections of it. The event types are the blueprint's 11 plus runtime extensions, all validated against backend/schemas/event.schema.json.
backend/agentteam/
contracts.py TeamPlan · Task · Event · Message · Artifact · Approval · Run
config/ YAML + JSON Schema · secret references (env: / keychain: / file:)
store/ EventStore · ArtifactStore · RunStore
providers/ anthropic_messages (official SDK) · openai_compatible_chat / ollama (httpx) · fake (tests only)
runtime/ PolicyEngine · MessageBus · ToolGateway · Sandbox · Checks · Web tools · AgentRunner · Planner · Scheduler · RunManager
api/ REST + SSE + static UI
frontend/ React + TypeScript (request · run · settings)
docs/ website, blueprint (spec, security, references), status
evals/ 40-case acceptance plan and coverage map
A real end-to-end run (2026-09-13) through the local Claude Code CLI, model reported by the provider: claude-opus-5. Request: Japanese launch page + three social-post drafts from a product description, stop before publishing.
| Plan | Master chose 1 builder task + 1 reviewer task and skipped the researcher; 8 recorded assumptions (no prices, no invented numbers, placeholder URLs) |
| Deliverables | index.html r1 (single file), posts.md r1, HANDOFF.md r1, final-report.md |
| Verification | 10 programmatic checks → all pass; reviewer verdict 6/6 pass + 4 optional findings delivered as a real message |
| Messages | builder → reviewer handoff, reviewer → builder finding |
| Usage | 39 model turns · 35 tool calls · $1.66 list-price equivalent · 18 min 37 s |
| Status | completed |
Evidence is committed unedited under docs/evidence/: the generated files, the final report and the 77-event JSONL log (run4-*). Earlier runs are there as well: run 1 ended partial and exposed a bug (a reviewer verifying two tasks had only its last verdict applied — fixed), runs 2–3 hit limits that were then tuned (docs/STATUS.md).
Also verified: 39 deterministic tests (cd backend && .venv/bin/python -m pytest -q) covering DAG validation, real delivery with a question/answer round trip, review → revise → re-review, budget reservation, approvals with hash/nonce, cancel → resume, fork, redaction, SSRF guard, sandbox write confinement, SSE cursor replay and event-schema conformance; headless-Chrome UI smoke with zero console errors.
Still open: one request and four runs is not a benchmark; plans vary between runs because the Master decides; prompts are original seeds. The Anthropic API and OpenAI-compatible drivers are implemented and unit-tested but have not had a live run yet. Full matrix: docs/STATUS.md. Security boundaries: SECURITY.md.
- dev.to: I built an AI team that ships real work — and shows you the conversation
- Zenn（日本語）: Bot 同士を会話させるのではなく、仕事の経緯がそのままチャットになるマルチエージェントを OSS で作った
- Launch threads on X: English · 日本語
The product was built from a written blueprint (docs/blueprint/): spec, implementation brief, security requirements, references, and 40 acceptance cases. Deliberate deviations (no Deep Agents/LangGraph dependency, extended event enum, refusal fallback off by default) are recorded in docs/IMPLEMENTATION_PLAN.md. Contributions: CONTRIBUTING.md.
依頼は一度。AI チームが作り、確かめ、成果物と経緯を残す。
Master が成果物と完了条件を決め、必要な Bot（Researcher / Builder / Reviewer）だけが実際に作業します。成果物、Bot 間の実メッセージ、時系列、最終報告は同じ実行記録（append-only Event Store）から表示され、台本の会話や固定の成功ログは使いません。
- 成果物を開くと、誰が作り、何を引き継ぎ、どの検証を通ったかまで辿れる
- 各 Bot の接続先・モデル・システムプロンプトを個別に上書きできる（共通設定を継承、手動固定あり）
- 停止・再開・分岐（Bot のモデルを変えて別案）・再生（LLM 呼出なし）・JSONL エクスポート
- 権限・予算・回数上限・承認はプロンプトではなく Runtime が強制
上の GIF はテスト用のスクリプト provider（LLM 呼出なし、画面に「FAKE PROVIDER」と表示）で実 UI と実 Runtime を動かしたものです。実 LLM での協働は 2026-09-13 にローカルの Claude Code CLI（claude-opus-5）で実施し、LP・投稿案・レビュー・最終報告まで completed で完走しました（証拠: docs/evidence/run4-*）。API キー不要で、claude にログイン済みなら agentteam probe → agentteam serve で始められます。詳細は docs/STATUS.md。
MIT. Bundled prompts and skills are original to this project (see docs/blueprint/REFERENCES.md).