Board
Board is the team workbench: a kanban of items the agent can execute, plus related work (tasks, goals, projects, hooks, blueprints). It is installed by default. Open Board in the sidebar.
Overview
- Kanban columns: Backlog → Active → Review → Done (archived separately)
- Create items, then drag between columns or reorder within a column (long-press on mobile)
- Rich description editor on create/edit; search and highlight from deep links (
?highlight=) - Auto-execution on assignment (toggle and provider in Board settings)
- Archive finished work; reopen from Archived items
- Done column hints auto-archive — clear finished work regularly
- Agents get Board MCP tools when the app is installed (default)
In the app
Board (/board)
- Open Board
- Add cards in a column (except archived)
- Drag items across columns or reorder within a column (on mobile: swipe between columns, long-press a card to pick it up)
- Use search to filter cards
- Open an item for rich description editing, assignment, and execution status
- Header links:
- Settings — auto-execution and provider
- Archived items —
/board/archive
If auto-execution is off, the board shows an Auto-execution disabled badge.
Board settings (/board/settings)
| Setting | Meaning |
|---|---|
| Enable board auto-execution | When on (default), assigned board items can run automatically for this team. When off, assignment does not trigger automatic agent execution. |
| Item execution provider | Which provider pack runs assigned items. Values are auto, openai, anthropic, gemini, and opencode; auto falls back to item meta, then team BYOK/env settings. |
Stored as team meta board.auto.enabled and board.auto.provider.
Item execution providers
Assigned items can run through different execution providers. Choose the provider in Board settings. auto falls back to item meta, then team BYOK/env settings.
Providers are implemented as Warp provider packs with the same lifecycle actions:
launch-taskget-task-statusadd-followupcancel-task
Some packs run through a hosted API. Local CLI packs (for example OpenCode) run on the user's desktop.
Local CLI execution
When a local CLI provider is selected:
- Working directory: each device has a local Item execution working directory in Board → Settings (default
~/Code). It is stored per device, not in team meta.JOAI_CODE_ROOTcan override it. - Repository tag: add
repo:<repository-directory>to the item tags (for example a repo directory namedmy-project). The directory must exist under the working directory and the repo must already have adevbranch. The tag is a convention read by the agent; it is not special server-side. - Agent flow: the agent reads the
repo:tag and calls the provider'slaunch-taskaction. When the pack accepts them, it passesREPO,RUN_ID,CALLBACK_URL, andCALLBACK_TOKEN. - Desktop execution: local CLI packs validate the repo, take a per-repo lock, switch to
dev, run the CLI, collect changed files andgit diff --stat, reject/undo commits and branch changes, restore the previous branch, and post the item execution callback. - Result:
completedmoves the item to Review.failedkeeps it Active and allows iteration (up to 5). - No automatic check script runs. Include any checks in the task or repository instructions if needed.
How to tag an item
Tags are explicit strings on the item. The repo: prefix is a convention read by the agent; it is not special server-side.
- Board UI: open the item and edit Tags.
- REST API:
POST /v1/itemsorPUT /v1/items/{id}with"tags": ["repo:<repository-directory>"]. - MCP:
create_itemorupdate_itemwithtags: ["repo:<repository-directory>"]. - Warps:
joai/item-createorjoai/item-updatewith thetagsinput. - Hashtags in the title cannot carry the colon; set the full tag explicitly.
Assigning agents
create_item assigns the calling agent by default. To assign a different agent — for example a tester — pass agents as an array of agent UUIDs. update_item replaces the existing agent assignments with agents; pass [] to clear them.
- MCP:
create_itemorupdate_itemwithagents: ["<agent-uuid>"]. - Agent UUIDs: use
joai_list_agentsor the Board UI. - Warps / REST:
joai/item-create/joai/item-updatewith theagentsJSON array. The REST API accepts the same field asagents.
Where related Board MCP concepts live in the UI
Board MCP covers more than the kanban. Product UIs:
| Concept | Where in the product |
|---|---|
| Items | Board columns |
| Tasks | Tasks / chat |
| Goals / reminders | Agent settings → Knowledge |
| Hooks | Agent settings → Automations → Hooks (not the Automations native app) |
| Blueprints | Agent settings → Identity (+ public Blueprints store) |
| Projects | Studio (/studio, project views) — not on the kanban |
| Updates (News) | News UI — MCP update tools are gated by Board |
Also see Blueprints, Alerts.
For agents (MCP)
Requires Board (default on most teams).
- Read tools use team view access
- Write tools use team manage access
Read: list_items, list_tasks, list_goals, list_projects, list_hooks, list_blueprints, list_reminders, list_updates, get_project, get_blueprint, get_update_stats
list_items defaults to items assigned to the calling agent. Pass scope=all to see the full team board.
Write: create_item, update_item, delete_item, create_task, toggle_task, delete_task, create_goal, update_goal_progress, create_project, delete_project, create_hook, toggle_hook, delete_hook, create_blueprint, update_blueprint, create_reminder, delete_reminder, create_update, update_update, delete_update
Live schemas: tools/list. See MCP.
Tips
- Keep Active lean so auto-execution stays predictable
- Use Review before Done when humans must approve agent output
- Turn off auto-execution when you want assignment without immediate runs