Headless task management for AI agents, CLIs, and chat-driven workflows.

ADHED is a self-hosted REST API for task management with a deterministic, server-side rules engine. No UI — API only. Designed for bots, agents, and scripts, not humans clicking buttons.

Linear, Notion, Todoist, and friends are UI-first. Their APIs are great, but the agents don't need the GUI.

ADHED is:

• API-first. Every feature is reachable over HTTP with curl. Interactive docs at /docs.
• Rules run on the server. You write a rule like "reject new issues without a priority if more than 3 are already in progress" once, and every agent / CLI / human writing through the API gets the enforcement for free.
• Deterministic. Same request → same behaviour. Broken rules fail loudly. Audit trail on every mutation.
• Yours. Self-hosted. Your data, your keys, your infra.

Built for the workflow where a chat agent is capturing tasks, a classifier is labelling them, a CLI is triaging, and maybe a human occasionally pokes the /docs page. The database is the single source of truth they all agree on.

$ ./setup.sh
  Your name: Jane
  Your email: jane@example.com
  Team name [Home]: Home
  ✅ ADHED is running!
  API Key: adhed_a1b2c3...
  URL:     http://localhost:8100

$ source .adhed-credentials
$ curl -s -X POST "$URL/api/v1/teams/$TEAM_ID/issues" \
    -H "Content-Type: application/json" \
    -H "X-API-Key: $API_KEY" \
    -H "X-User-Id: $USER_ID" \
    -d '{"title":"Call the dentist","priority":2}' | jq .data.title
"Call the dentist"

$ curl -s "$URL/api/v1/teams/$TEAM_ID/summary" \
    -H "X-API-Key: $API_KEY" \
    -H "X-User-Id: $USER_ID" | jq .data.triage_count
1

git clone https://github.com/dimatosj/adhed.git
cd adhed
claude

Claude walks through Docker setup, prompts for your name/email/team, starts services, saves credentials, and points you at /docs.

git clone https://github.com/dimatosj/adhed.git
cd adhed
./setup.sh

Requires Docker + curl. Everything else is handled.

Services started: API server, Postgres, and a scheduler for recurring tasks.

• Issues with subtasks, labels, projects, priorities, due dates, estimates, custom fields
• Six-state workflow (triage → backlog → todo → in progress → done → canceled) with enforced valid transitions
• Fragments — typed notes (person, place, credential, memory, idea, resource, journal) with topics, domains, entities, and full-text search. Structured context for agents, not just tasks.
• Fragment links — connect fragments to issues, projects, or other entities with typed relationships
• Recurring tasks — cron or interval schedules that automatically spawn issues into backlog. Title templates with {date} substitution, configurable issue defaults, pause/resume.
• Rules engine — trigger → condition → action, evaluated server-side on every write. Reject, auto-label, set fields, notify, add comments.
• Full-text search on issues and fragments (Postgres TSVECTOR)
• Batch operations — create and update many issues at once
• Audit trail on every mutation across every entity
• Summary dashboard — triage count, overdue, due soon, stalled projects, waiting-for
• Role-based access (OWNER / ADMIN / MEMBER) with MEMBER audit self-filtering
• Multi-team, team-level API keys (SHA-256 hashed at rest)

Interactive docs: http://localhost:8100/docs after setup.

See the full API Reference.

Write a rule once, and it runs on every write. Examples:

• "Auto-label any issue whose title contains 'dentist' or 'gym' with health."
• "Reject moving an issue from triage unless priority is set."
• "If more than 3 issues are in progress, reject creating another."
• "When an issue is assigned, notify the assignee."

See Rules Engine for the full DSL.

• NanoClaw / OpenClaw — drop-in skill for Claude-in-container workflows
• Any HTTP client — it's REST with JSON in / JSON out

• Getting Started — setup walkthrough
• API Reference — every endpoint, parameter, response
• State Machine — the six state categories and transitions
• Rules Engine — conditions, actions, examples
• Fragments — typed notes for agent context
• Recurrences — recurring task schedules
• Integration Guide — building clients
• Architecture — the three-layer design
• Deployment — running beyond localhost

Bug reports, security reports, and PRs welcome. See CONTRIBUTING.md and SECURITY.md.

MIT.