The AI agent command center

Run your AI agents like one team.

Dispatch is a local-first command center for multi-agent coding. Assign work to Claude, Gemini, OpenAI, local models, or your own CLIs — then watch them plan, execute, hand off, and ship from one workflow.

Bring your own tools. No token markup. Not another chat window.

See how it works

Native macOS app — coming soon

The Dispatch desktop app: four AI agents working in a 3D office, with a live task board, system panel, and a streaming console.

Works withClaude Code·Gemini CLI·OpenAI-compatible endpoints·local models·custom shell commands

The problem

Multiple agents shouldn't mean ten terminals.

When one agent isn't enough, the work scatters — terminals, tabs, sessions, no memory of who's doing what or what it spent. Dispatch gives every agent a desk, a task, and a visible queue. Watch the floor at a glance, then drill into the logs when you need them.

How it works

From your tools to a working floor.

Detect

Dispatch finds the AI tools already on your machine — Claude Code, plus any Anthropic endpoint or OpenAI-compatible API.

Create agents

Give each one a name, a role, a model, and a project folder. They show up at their desks, ready for work.

Assign tasks

Drop work on the board and pick an agent. Create-and-run to start now, or queue it in the backlog.

Watch them ship

Agents run, hand work off, and move cards to Done — with live logs, status, and cost in the open.

Model-agnostic

One task. Many models. One team.

Dispatch isn't tied to a single model. Route each step to whatever's best, and let them work together — Claude plans, Gemini scans, OpenAI reviews, a local model handles the private parts.

ClaudePlanning, refactors, and implementation
GeminiLarge-context repo scans and broad analysis
OpenAIStructured output, review, and tests
Local modelsPrivate or simple work, at zero cost
Your own CLIsWhatever agent setup you already run

Example

Assign Claude to implement auth, Gemini to scan the repo, OpenAI to write the tests, and a local model to summarize the handoff. Dispatch keeps every step on one board.

Workflow protocol

Agents don't just run. They hand off.

Every task moves through the same pipeline, and each agent emits a standard result so the next one picks up exactly where the last left off.

TaskPlanAssignExecuteHandoffReviewApproveDone

handoff.json

{
  "status": "needs_review",
  "summary": "Implemented login form validation",
  "files_changed": ["src/Login.tsx", "src/auth/schema.ts"],
  "next_suggested_agent": "security-reviewer",
  "risks": ["Needs a manual test with real credentials"]
}

Native for macOS

Coming soon for macOS.

Dispatch is a real Mac app — your agents and your CLIs, running locally on your own machine. Mac App Store submission is planned; request a demo to get into the private beta first.

Run your AI workforce from one command center.

Dispatch is opening its macOS beta to people running more than one agent at a time. Bring your own tools and get early access.

Built in a hotel the night before a wedding — because agents needed a dispatch board, not another chat window.