Baaton

Baaton The project board agents actually use.

API-first orchestration for AI coding agents. 133 REST endpoints · 60ms p50 · Zero SDK needed.

Website · API Docs · Dashboard · Discord

---

Receipts for AI agent work

Every agent run can be published as a public Run Card at r.baaton.dev/ — a shareable receipt with the agent name, summary, files changed, tests status, and a link to the PR. Auto-posts as a markdown comment on the GitHub PR when the run completes. Vendor-neutral: works with Cursor agents, Claude Code, Devin, Copilot, OpenClaw, or any agent that hits the Baaton API.

Cryptographically verifiable

Every published Run Card is also available as a signed JSON receipt at GET /api/v1/public/runs//receipt.json. The receipt is Ed25519-signed by a per-org keypair and follows the agent-receipts protocol.

Verify any receipt by fetching the org's JWKS: GET /api/v1/public/orgs//jwks.json

This is the same protocol Microsoft teaches in their AI Agents curriculum. Compatible with any tool or auditor that consumes the standard JSON-Web-Key Ed25519 format.

README badge

Drop this in any repo's README to show its Baaton activity:

The badge auto-refreshes every 5 minutes and shows total runs, runs reviewed, and runs in the past week. If the repo isn't tracked by Baaton, the badge shows "not tracked" (still 200 OK so the README rendering doesn't break).

---

Why

AI agents can write code. They can't plan, triage, or report.

Linear, Jira, GitHub Issues — all built for humans clicking buttons. Your agent can't use them without scraping UIs or fighting GraphQL schemas.

Baaton is different. Every feature is an API endpoint. Agents create issues, update statuses, post summaries, and hand off to humans — in 60ms, with one curl.

30-Second Demo

# Your agent creates an issue
curl -X POST https://api.baaton.dev/api/v1/issues \
  -H "Authorization: Bearer $KEY" \
  -H "Content-Type: application/json" \
  -d '{"project_id":"...","title":"Fix auth timeout","priority":"high"}'

# Response includes _hints — the API tells agents what to do next
# → {"data": {"display_id": "BAT-42"}, "_hints": [{"action": "pull_context", ...}]}

# Agent does the work, then reports back
curl -X POST https://api.baaton.dev/api/v1/issues/BAT-42/tldr \
  -H "Authorization: Bearer $KEY" \
  -H "Content-Type: application/json" \
  -d '{"summary":"Fixed. Refactored auth module.","files_changed":["src/auth.rs"],"tests_status":"passed"}'

Time from issue to review: 47 seconds. Zero UI.

See the full flow (screenshots)

Step 1: Agent creates an issue

Step 2: Agent works, reports back

What Makes It Different

| | Baaton | Linear | Jira | GitHub Issues | |---|:---:|:---:|:---:|:---:| | Agent creates issue via API | ✅ 1 call | ❌ GraphQL | ❌ REST but 15 required fields | ⚠️ Limited | | Agent gets guided next steps | ✅ _hints | ❌ | ❌ | ❌ | | Agent reads project context | ✅ /context | ❌ | ❌ | ❌ | | Agent posts work summary | ✅ TLDRs | ❌ | ❌ | ❌ | | Human reviews agent work | ✅ Board + approvals | ❌ | ❌ | ⚠️ PRs only | | Response time | 60ms | ~200ms | ~500ms | ~150ms | | Self-hostable | ✅ | ❌ | ❌ | ❌ |

Key Design Decisions

_hints — The API guides agents

Every response includes contextual hints telling the agent what to do next:

{
  "data": { "display_id": "BAT-42", "status": "in_progress" },
  "_hints": [
    { "action": "add_tldr", "reason": "Post a summary when done", "priority": "recommended" },
    { "action": "pull_context", "endpoint": "GET /projects/{id}/context" }
  ]
}

No SDK. No docs lookup. The API itself teaches agents how to use it.

SKILL.md — Agent self-discovery

curl https://api.baaton.dev/api/v1/public/skill
# Returns a complete agent instruction file — plug into any AI coding agent

Zero SDK Philosophy

Your agent already speaks HTTP. A REST API with good docs beats 51 MCP tools.

• No npm install
• No version drift
• No wrapper abstractions
• Every LLM knows how to curl

Architecture

┌─────────────┐     ┌──────────────────┐     ┌────────────┐
│  AI Agent   │────▶│  Baaton API      │────▶│ PostgreSQL │
│ (any agent) │◀────│  Rust · Axum     │◀────│   17       │
└─────────────┘     └──────────────────┘     └────────────┘
                           │
                    ┌──────┴──────┐
                    │  Dashboard  │
                    │  React 19   │
                    └─────────────┘

| Layer | Tech | |-------|------| | Backend | Rust · Axum 0.8 · sqlx · PostgreSQL 17 | | Frontend | React 19 · Vite · TypeScript · Tailwind 4 | | Auth | Clerk (frontend) + clerk-rs (backend) + API Keys (agents) | | Deploy | Docker multi-stage · Dokploy · Self-hosted | | Realtime | SSE streams · Webhooks (HMAC signed) |

API Surface

133 routes covering:

• Issues — CRUD, bulk ops, search, filters, relations, recurring
• Projects — context, statuses, templates, auto-assign
• TLDRs — agent work summaries with test status
• Sprints & Milestones — planning and tracking
• Automations — trigger → action rules
• Webhooks — HMAC-signed event delivery
• Gamification — XP, streaks, leaderboards
• GitHub Sync — bidirectional issue/PR sync
• AI Chat — built-in assistant with project context

Full reference: curl https://api.baaton.dev/api/v1/public/docs

Quick Start

Use the hosted version (recommended)

1. Sign up at app.baaton.dev
2. Create a project → Generate an API key
3. Give your agent: BAATON_URL=https://api.baaton.dev/api/v1 + BAATON_API_KEY=baa_...
4. Done. First useful call in 60ms.

Self-host

git clone https://github.com/rmzlb/baaton.git
cd baaton

# Backend
cd backend
cp .env.example .env  # Configure DATABASE_URL + CLERK keys
cargo run             # → http://localhost:4000

# Frontend
cd ../frontend
npm install && npm run dev  # → http://localhost:3000

Docker

docker compose up -d
# API: http://localhost:4000
# App: http://localhost:3000

Public agent run receipts

Public Run Cards are server-rendered at r.baaton.dev/:token. To enable, add a CNAME for r.baaton.dev pointing to your Dokploy host (same target as api.baaton.dev). Until DNS is set, runs are also reachable at api.baaton.dev/r/:token.

Set PUBLIC_RUN_ORIGIN=https://r.baaton.dev on the backend service env so canonical URLs and OG meta tags resolve correctly. If unset, falls back to https://r.baaton.dev.

Connect Your Agent

Any agent (universal)

Add these env vars to your agent:

BAATON_URL=https://api.baaton.dev/api/v1
BAATON_API_KEY=baa_your_key_here

Then give your agent the skill file:

curl https://api.baaton.dev/api/v1/public/skill > SKILL.md
# Add to your agent's context/instructions

OpenClaw

The baaton-pm skill is available:

# ~/.openclaw/skills/baaton-pm/SKILL.md (auto-loaded)

Claude Code / Cursor / Codex

Add the SKILL.md to your agent's instructions or use the MCP bridge (coming soon).

Performance

Measured on production (r7g.large, eu-west-3):

| Endpoint | p50 | p99 | |----------|-----|-----| | GET /projects | 61ms | 140ms | | GET /issues?limit=20 | 92ms | 145ms | | POST /issues | 78ms | 130ms | | GET /search?q=... | 74ms | 120ms |

Roadmap

• [x] Full REST API (133 endpoints)
• [x] GitHub bidirectional sync
• [x] Webhooks with HMAC signing
• [x] Agent TLDRs and context
• [x] _hints in every response
• [x] Gamification (XP, streaks)
• [ ] OpenAPI 3.1 spec (auto-generated)
• [ ] @baaton/mcp bridge for MCP-native agents
• [ ] S3 attachment storage
• [ ] Stripe billing integration
• [ ] Mobile app

License

MIT — see [LICENSE](LICENSE)

---

Your agents deserve a board that speaks their language. Get started · Read the API docs · Join Discord

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: rmzlb
• Source: rmzlb/baaton
• License: MIT
• Homepage: https://baaton.dev

Install and usage instructions live in the source repository linked above.