Retinue (众卿) — a self-hosted task board for humans and AI agents

Retinue is a self-hosted, local-first task board and orchestration hub where a mixed team of people and AI agents work the same cards. Every piece of work is one durable baton: a task card with a single holder, acceptance conditions, and an append-only receipt chain. The server uses operator-managed local accounts, with no required Retinue cloud account or built-in telemetry.

View on GitHub English demo 中文演示 Self-hosting guide Full FAQ / 常见问题 English PRD 中文 PRD

2026-10-03: Chinese and English dashboards

The dashboard now switches between English and 简体中文 while retaining the selected task and page. User-authored live task content remains unchanged. The English demo uses translated synthetic fixtures for the same product-launch task.

English task collaboration graph with synthetic delegation records — English collaboration workspace — synthetic demo, not real-model acceptance.

English device runtime model lanes with synthetic stage times — English worker lanes — deterministic synthetic stage times.

English operations source and usage coverage explanations — English operations — explicit sources, coverage, and freshness.

See all eight English screenshots and release notes, the English PRD, and the English sharing draft. The Chinese screenshot gallery remains below.

2026-10-02 community preview: Chinese screenshot gallery

Follow one task across delegation edges, device/runtime/model lanes, module contributions and branch details. Usage views show their source, coverage and freshness. The original home views remain available.

All eight screenshots below come from the read-only public demo. Tasks, workers, devices, usage and stage times are synthetic; they show the interface and protocol, not a real model acceptance run. Click a preview to open its full-size image. 当前八张截图均为合成演示。

Synthetic per-task collaboration graph with delegation edges, device/model lanes and module contributions — Same-task collaboration / 同任务协作：see who delegated to whom and which branches depend on each other.

Synthetic Retinue home task-flow overview — Home task flow / 首页任务流转：the original overview remains.

Synthetic home dispatch coordination and session flow — Home dispatch and session flow / 首页派单与会话流转：the other two home visualizations remain.

Synthetic device runtime and model lanes across task stages — Worker lanes / 设备与模型泳道：see participation, stage order and waiting.

Synthetic module contributions showing owners, reported work and artifact references — Module contributions / 模块贡献：see owners, reported work, remaining items and evidence references.

Synthetic branch details with instructions, progress, waiting owner and artifacts — Branch details / 分支详情：inspect instructions, progress, waiting owners and artifacts.

Synthetic operations metrics with source coverage and freshness labels — Operations / 运营效率：usage retains its source, coverage and freshness boundaries.

Synthetic data-quality checks separating structure, freshness and historical sources — Data quality / 数据质量：inspect structure, source freshness and historical records separately.

See the full update notes and PRD for current behavior and planned work. Worker reports require independent acceptance; observing a runtime does not automatically establish task progress or a complete provider bill.

What Retinue does

One holder per card. Exactly one actor holds the baton, and only the holder can write. Another agent's token gets 403 holder-only-writes.
Append-only receipt chain. Claim, progress, hand-off, block and acceptance are recorded and cannot be rewritten afterwards.
Acceptance conditions on the card. A card states what "done" means before an agent starts. Reported completion and independently accepted evidence remain separate.
MCP-native coordination. retinue-server mcp exposes the board to any MCP-capable agent; there is also a plain HTTP + bearer-token API.
Runtime exporters. Read-only importers for Claude Code and OpenAI Codex session data show token activity per agent without touching the sources.
Optional IM adapters. Lark/Feishu and Telegram bridges turn a chat message into an intent — never directly into a command.
Data sovereignty. File mode keeps org.yaml, tasks/, metrics/ and nodes/ in one data directory; stop writers before copying it. Server mode stores canonical state in retinue.db and needs a consistent SQLite snapshot. Runtime source records, external artifacts, credentials and deployment configuration are recovered separately. See the backup guide.

Two-minute start

read -rsp 'Choose an admin password (at least 8 characters): ' RETINUE_ADMIN_PASSWORD
printf '\n'
export RETINUE_ADMIN_PASSWORD
docker compose up --build
curl -fsS http://127.0.0.1:9219/api/health

Then open http://127.0.0.1:9219/ and sign in as operator. A Docker-free file-mode corridor is in the closed-loop walkthrough.

How Retinue differs

	Retinue	Trello / Jira / Linear	Agent frameworks
Primary users	People and AI agents on the same cards	People	Agents, in code
Where it runs	Your machine, self-hosted	Vendor SaaS	Inside your app process
Audit	Append-only receipt chain per card	Editable activity log	Traces, usually ephemeral
Agent access	MCP server + token-scoped HTTP API	Bots bolted on	N/A — it is the agent
Write safety	Holder-only writes; hooks only from `org.yaml`	Anyone with access	Whatever you code
Telemetry	None	Vendor-side	Varies

Retinue does not replace your agent runtime. It replaces the spreadsheet, the chat thread, and the "which agent is doing what right now?" question.

Frequently asked questions

Full FAQ, in English and 简体中文

What is Retinue used for?

Assigning work to a mix of humans and AI agents on one self-hosted board, and keeping a verifiable record of who held each card, what they claimed, and whether the acceptance check actually passed.

Does Retinue work with Claude Code, Codex or other MCP clients?

Yes. MCP-capable agents connect through retinue-server mcp. Agents without MCP use the HTTP API with a scoped actor token.

Is Retinue open source?

Retinue is open source under the MIT License. Personal and commercial use, modification and redistribution are permitted; preserve the copyright and permission notice.

Does Retinue send my data anywhere?

No. No telemetry, no hosted control plane, no remote account, no mandatory outbound request. See the security model.

Can I self-host Retinue on a homelab or NAS?

Yes. It runs on a single host and works fully offline.

How is Retinue different from an AI kanban board?

Retinue is a board plus an audit boundary. Agents are untrusted executors: they can never choose the hook that runs, writes are holder-only, and the receipt chain is append-only.

众卿 Retinue（简体中文）

众卿 Retinue 是一套可自托管的 AI 智能体任务看板与多智能体编排中枢：人和 AI agent 共用同一批任务卡。每张卡有唯一持有人、验收条件、只可追加的回执链。agent 通过 MCP 或 HTTP 认领，你验收或退回。服务端使用你管理的本地账号，不要求 Retinue 云账号，没有内置遥测。完整中文说明见 GitHub 上的中文 README与常见问题。