Jarvis Kanban Dashboard

A lightweight kanban board for managing tasks for Jarvis/Clawdbot — now with Projects + basic task hierarchy (sub-tickets).

• Frontend: Vue 2 + Vuetify 2 (CDN, no build step)
• Backend: Node.js (Express) + SQLite
• Drag & drop: SortableJS via vuedraggable

Columns

• Ideas
• To do
• In Progress
• Review
• Testing
• Done

Jarvis should only pick up tasks in To do and In Progress.

Run

cd /home/clawdbot/clawd/kanban-dashboard
npm install
HOST=0.0.0.0 PORT=3000 npm start

Open: http://<server-ip>:3000/

API (summary)

• GET /api/health

• GET /api/columns

• GET /api/ai/health (shows whether LLM is configured/enabled)

• POST /api/ai/tasks/merge (AI create: enhance + optional breakdown into sub-tickets; returns plan + created tasks when applied)

• POST /api/ai/projects/:id/evolve (AI: propose new tickets based on current state)

• POST /api/ai/projects/:id/remove-concept (AI: remove/unmerge a concept from the hierarchy)

• POST /api/ai/projects/:id/init (AI: generate/overwrite project scope from a repo path; optional seed tickets)

• GET /api/tasks/:id/ancestors (optional ?include_self=1)

• GET /api/tasks/:id/descendants (optional ?mode=flat|tree&include_self=1)

• GET /api/tasks/:id/context

• POST /api/tasks/:id/next (select next leaf sub-ticket; LLM-assisted when enabled, heuristic fallback)

• GET /api/projects

• POST /api/projects

• GET /api/projects/:id

• PATCH /api/projects/:id

• DELETE /api/projects/:id (reassigns tickets to default project, then deletes)

• GET /api/tasks (optional ?status=todo&project_id=default&parent_id=<id>)

• POST /api/tasks (supports project_id + parent_id)

• GET /api/tasks/:id

• PATCH /api/tasks/:id (supports status, project_id, parent_id)

• POST /api/tasks/:id/comments

• POST /api/tasks/:id/updates (append-only progress log)

• GET /api/tasks/:id/updates?limit=5

Database

SQLite file: kanban-dashboard/data/kanban.db

Tables:

• projects (project scope/context lives here)
• tasks
• automation_tasks (recurring/scheduled; now also linked to projects)
• comments
• progress_updates
• llm_plans (stores validated AI "plans" and whether they were applied)
• llm_runs (stores raw LLM request/response/error logs for auditing)

LLM / AI (Phase 3)

Phase 3 adds:

• llm_runs table (logs LLM requests/responses/errors)
• /api/ai/health
• POST /api/tasks/:id/next (LLM-assisted next-leaf selection)

Environment variables

The backend uses an OpenAI-compatible Chat Completions endpoint (works with LiteLLM Proxy, OpenRouter direct, etc).

Common setup (LiteLLM proxy):

LLM_ENABLED=1
LLM_BASE_URL=http://127.0.0.1:4000
LLM_API_KEY=sk-...              # optional depending on your proxy
LLM_MODEL=openrouter/google/gemini-3-pro-preview

Optional tuning:

LLM_TIMEOUT_MS=45000
LLM_TEMPERATURE=0

AI create (Phase 4)

POST /api/ai/tasks/merge accepts:

{
  "project_id": "default",
  "parent_id": null,
  "status": "todo",
  "text": "Add OAuth login...\n\nMore details...",
  "breakdown": true,
  "max_depth": 3,
  "apply": true
}

Response (when apply=true) returns:

• plan_id
• llm_run_id
• root_task
• created_tasks[]

Guardrails / safety (Phase 5)

Optional API key for AI endpoints

If set, all /api/ai/* endpoints require X-API-Key: <value>.

AI_API_KEY=changeme

Rate limiting (in-memory)

Simple per-IP rate limiting for AI/LLM calls.

AI_RATE_LIMIT_PER_MIN=60

Project init repo scanning safety

By default, init can only read paths under the server working directory. You can allowlist additional roots:

AI_INIT_ALLOWED_ROOTS=/home/clawdbot,/mnt/data
AI_INIT_MAX_FILES=400
AI_INIT_MAX_FILE_BYTES=64000
AI_INIT_MAX_TOTAL_BYTES=600000

To allow any path (dangerous on shared machines):

AI_INIT_ALLOW_ANY=1