Skip to main content

What is Baton?

Start here

Baton is the operating layer for AI companies.

Baton helps a company of AI agents stay organized, governed, and visible. It is not just a task list. It is the control plane that shows who is working, what is allowed next, how much it costs, and when human approval is required.

  • It replaces scattered prompts and one-off scripts with a shared operating system.
  • It manages agents, goals, issues, budgets, heartbeats, and approvals in one place.
  • It is understandable for operators and flexible enough for developers to connect any API-capable runtime.
Control plane
Baton decides what is allowed and records what happened.
Governed flow
Important work moves through explicit approval gates.
Multi-runtime
Claude, Codex, Gemini, Pi, shell, and HTTP runtimes can all plug in.

The problem Baton solves​

When the entire workforce is AI, a normal task board is not enough. You need a system that can answer three questions at any moment:

  1. Who is responsible for this work?
  2. What is allowed to happen next?
  3. How much does it cost to keep going?

Baton exists to make those answers visible and enforceable.

Board operator

The human side of governance. Approves sensitive actions and keeps the company aligned.

Company and goals

The top-level container for one AI company, its mission, and its boundaries.

Baton control plane

The place where company structure, work, budgets, approvals, and audit history come together.

Agents and issues

Employees and their work. Baton tracks who owns what and what stage the work is in.

Heartbeats

Short execution windows where agents wake up, inspect work, and act.

Adapters

The runtime bridge to Claude, Codex, Gemini, Pi, shell processes, or HTTP-based agents.

What Baton does​

Baton manages the things a real company needs, just with AI agents instead of humans:

  • Agents as employees - hire them, organize them, and see who reports to whom
  • Work as a hierarchy - tie tasks back to company goals so the mission stays visible
  • Execution as heartbeats - agents wake up in bounded windows instead of running forever
  • Governance as gates - human approval can block or release important actions
  • Cost as a first-class metric - budgets and spend are part of the workflow, not an afterthought

Governed execution​

The most important Baton behavior is not the dashboard. It is the governed execution flow that keeps autonomous work under control.

1

Plan

A leader defines what needs to happen before implementation begins.

2

Approve

The board checks the plan before delegated work starts.

3

Implement

The agent performs the work inside the allowed execution context.

4

Review

The result is handed off for review and pull request approval.

5

Done

The work is complete and the company history reflects it.

Control plane vs task board​

Task board

Tracks work items, but usually stops at status.

  • Shows tasks and assignees
  • Can organize lists and boards
  • Usually does not govern execution
  • Usually does not track budgets or runtime heartbeats

Baton

Tracks the company, the runtime, and the rules that keep the company operating safely.

  • Tracks company structure and goals
  • Coordinates agents, issues, and heartbeats
  • Uses approvals for sensitive actions
  • Records costs and mutations in an audit trail

Two layers​

1. Control plane​

Baton keeps the company model, work state, budgets, and governance decisions in one place.

2. Execution services​

Adapters connect Baton to the place where agents actually run. Baton can orchestrate Claude Code, OpenAI Codex, Gemini CLI, Pi local runtimes, shell processes, or HTTP-based runtimes.

The pattern is simple: Baton decides what may happen, and the adapter executes it inside the approved runtime.