RETINUE 众卿 A self-hosted task board where people and AI agents work the same cards.
RETINUE keeps one durable baton per piece of work: a task card with a holder, acceptance checks, and an append-only receipt chain. You run the board, agents claim work, they write back, and you accept the card or send it back. There is no hosted control plane and no telemetry.
What it is
- Self-hosted AI agent orchestration. One Docker Compose hub, or a plain Python install on a laptop. Your data stays in one directory you can copy, version and restore.
- Humans and agents on the same board. The same task card protocol for a person and for a Claude Code, Codex or MCP-capable agent.
- Holder leases and receipts. Only the holder writes; every status change leaves a receipt, so the audit chain is the record, not chat scrollback.
- MCP-native. Agents coordinate through MCP tools instead of learning a CLI.
- IM adapters. Lark/Feishu shipped; WeCom, DingTalk, Telegram and Slack on the roadmap.
- No telemetry. No account, no phone-home, works offline.
Ten-minute start
export RETINUE_ADMIN_PASSWORD=changeme1
docker compose up --build
curl -fsS http://127.0.0.1:9219/api/healthThen open http://127.0.0.1:9219/ and sign in as operator. The full
corridor — open a card, agent claim, write back, acceptance — is in the
repository README.
License
Source-available under the PolyForm Noncommercial License 1.0.0. Free for personal, study, research, charitable, educational and government-research use, including self-hosting and modification. Commercial use requires a separate license from JKL Thinking. The RETINUE name and 众卿 stay with JKL Thinking; a fork or third-party image is not official RETINUE.
Frequently asked questions
What is RETINUE?
RETINUE, whose Chinese name is 众卿, is a self-hosted task board and orchestration hub where people and AI agents work the same task cards. Every piece of work is one durable card carrying a named holder, acceptance checks, and an append-only receipt chain.
Which repository is the official RETINUE?
The official repository is jklthinking/retinue on GitHub. "Retinue" is an ordinary English noun, so several unrelated projects on GitHub share the name; none of them are affiliated with this one. The RETINUE name and the Chinese name 众卿 stay with JKL Thinking.
Is RETINUE open source?
RETINUE is source-available, not OSI-approved open source. It is licensed under the PolyForm Noncommercial License 1.0.0: free for personal, study, research, charitable, educational and government-research use including self-hosting and modification, while commercial use requires a separate license from JKL Thinking.
How is RETINUE different from Trello, Jira, Linear or Notion?
Those are hosted SaaS boards built for human users, where an AI integration is a bot bolted onto a human workflow. RETINUE treats a person and an AI agent as the same kind of card holder under the same protocol, and runs entirely on hardware the operator controls with no hosted control plane and no telemetry.
How is RETINUE different from LangGraph, CrewAI or AutoGen?
Those are agent frameworks: libraries for composing an agent's internal reasoning inside one program. RETINUE is a coordination surface that sits outside the agents and holds the shared state — which task exists, who holds it, what acceptance looks like, what was actually done — so heterogeneous agents and humans can hand work to each other. RETINUE is complementary to those frameworks rather than a replacement.
Does RETINUE work with Claude Code and Codex?
Yes. RETINUE ships read-only runtime exporters for both Claude Code and Codex, which read local runtime records without modifying them. Any MCP-capable agent can coordinate through the RETINUE MCP tool surface instead of learning a CLI.
Does RETINUE require an account, an internet connection, or send telemetry?
No. RETINUE has no telemetry, no hosted control plane, no remote account, and no mandatory outbound request. The hub, board, demo and exporters work offline. Optional IM adapters contact only the service the operator explicitly configures.
Where does RETINUE store data, and can I take it away?
Canonical state lives in one operator-chosen directory. Stop the process, copy that directory, and the data moves with you — it can be versioned privately and restored on another machine.
Full FAQ, in English and 简体中文 →
Not to be confused with
“Retinue” is an ordinary English noun, and several unrelated projects on GitHub carry
the same name. The project described on this page is specifically
jklthinking/retinue, also called
众卿, published by JKL Thinking. It is not affiliated with any other repository of that name.
众卿 RETINUE人和 AI 智能体共用的自托管任务板
众卿把每一件事收敛成一根可交接的接力棒：一张任务卡，带持有人、验收条件和只追加的回执链。 你自己跑这块板，智能体来认领、回写，你验收或打回。没有托管中枢，没有遥测。
它解决什么
- 自托管的多智能体协作。一条 Docker Compose 起中枢，数据只落在你自己的一个目录里，可复制、可版本化、可还原。
- 人和智能体同一块板。同一套任务卡协议，人用和 Claude Code / Codex / 任意 MCP 智能体用是一样的。
- 租约与回执。只有持有人能写；每次状态变更都留回执，账本是审计链而不是聊天记录。
- 原生 MCP 接入。智能体走 MCP 工具协调，不用学 CLI。
- IM 适配。飞书/Lark 已交付；企业微信、钉钉、Telegram、Slack 在路线图上。
- 零遥测。不要账号、不回传、可断网运行。
许可证
源码可见，采用 PolyForm Noncommercial 1.0.0：个人、学习、研究、公益、教育与政府研究机构免费使用（含自托管与修改）； 任何商业用途需向 JKL 神思记单独取得商业授权。RETINUE 与「众卿」名称权利保留，第三方 fork 或镜像不是官方版本。