Limited private beta

Your coding agents.One calm board.

Run Claude Code, Codex and Antigravity in isolated worktrees from one focused board.

Works withClaude Code logoClaude CodeCodex logoCodexAntigravity logoAntigravityCursor and OpenCode, planned

Product tour

From task to reviewed change.

See the real CatBoard app in three focused views.

01 · Create a task

Starting parallel agents is easy. Keeping their work separate, visible and reviewable is the hard part.

Runs disappear into terminal windows.

Status, output and finished work end up scattered across separate sessions.

Shared checkouts create avoidable conflicts.

Two agents can edit the same working tree before either one is ready to review.

Landing the result interrupts your own work.

Switching branches and untangling changes pulls you away from the task you were doing.

Features

Built for parallel work

01 / Isolated

Isolated worktrees

Every task starts on its own branch and directory, so parallel agents never edit the same working tree. Blocked tasks wait for their dependencies. Opt-in batches can continue when the next task is ready.

02 / Observable

Structured transcript

Agents run with JSON event streams, so turns, tool calls and results render as real UI. Track reported token usage and available cost totals by repository.

03 / Reviewable

Review the work, not terminal scrollback

Read the summary, inspect the diff and merge from the same focused view without switching branches in your main checkout.

More capabilities

Stay in flow as the work grows.

Command paletteLight and dark themesRun historyDiff viewResume a conversationNative notificationsOpen in VS Code, Cursor or TerminalTicket batches

CatBoard Skill + MCP

Turn agent conversations into an actionable board.

Install the CatBoard skill and MCP server once. Then your coding agent can understand board context and propose structured task batches for your review.

Agent conversation

Claude Code

/catboard

Add rate limiting to the public API.

I split this into three tasks with two dependencies. Ready to send the proposal to CatBoard.

What happens here

The CatBoard skill gives your agent the workflow for turning a conversation into clear tasks, dependencies and repository context.

Proposed task batch

API rate limiting

Awaiting approval
01

Add token bucket store

API · Codex

02

Wire limiter middleware

Needs 01

03

Return rate limit headers

Needs 02

MCPSends the proposal to CatBoard
Approve 3 tasks

01 / Skill

Shapes the work

Turns the discussion into tasks with dependencies and handoff context.

02 / MCP

Connects to the board

Reads relevant board context and submits the proposed batch to CatBoard.

03 / You

Approve before creation

Review tasks, dependencies and repository targets before anything is added.

Local by default

Local orchestration, by default.

CatBoard orchestrates the tools already installed on your Mac and stores its board data locally. It does not host or upload your repositories.

01

No CatBoard account for free use

02

Analytics off by default

03

CatBoard never uploads your repositories

04

Runs and history stay local

Limited private beta

Put your agents on the board.

Selected beta users get three months of Pro with unlimited agent runs, at no cost.

3 months of Pro included

Unlimited agent runs during your Pro access

Unlimited repositories and task cards

All three supported agents

Signed and notarized macOS build

Access is limited. We will email you if a beta spot is available.

Which agent do you use most?

FAQ

Questions

  • You need each agent CLI installed and authenticated. Provider subscriptions and usage charges remain with that provider.

  • CatBoard does not upload or host your repositories. The agent CLI you choose may send prompts, files or context to its own provider according to that provider's terms.

  • macOS 14 or newer, git 2.38 or newer (for merge-tree --write-tree).

  • CatBoard does not check out the task branch or rewrite files in your main working tree. When a merge succeeds, it advances the selected base branch through Git refs.

  • The worktree stays where it is. Retry the task, edit the prompt and retry, or open the worktree in a terminal and finish it by hand.

  • Anonymous product analytics are optional and off by default. When enabled, they never include prompts, repository contents or agent output.