Your agents can do more.A lot more.

CuttleCode is an AI coding harness that unlocks the full potential of your agents. Bring Claude Code, Codex or whichever providers you already use, and get everything they are missing.

Download (Beta)Explore the product

Claude Code, Codex, and more

Bring your own
subscription.

Plus OpenCode, Amp, Ollama, llama.cpp, and custom agents. Provider accounts and terms apply.

Use the coding agents you already subscribe to. Spawn sessions, give agents roles, and create workflows from simple handoffs to elaborate coordination across sessions.

Agents in control

Superpowers in every session.

CuttleCode gives your agents capabilities other coding tools lack. Reduce development friction by allowing your agents to drive the harness themselves. If you can do it, they can do it.

Under the hood, every session has the cuttle CLI, so it drives CuttleCode the same way you can.

Session · Procedural dungeonsReady

Pipelines

Run a workflow across multiple agents.

Chain agents together like an assembly line. Each agent does one job and hands its work to the next, so a whole workflow runs start to finish without you. Start it yourself or put it on a schedule.

Overnight bug sweepReady
Example: an overnight bug sweep. One agent finds and fixes bugs, another tests each fix, a third decides what’s safe to merge, and a summary lands in Discord.

Autonomy

Built-in agent loops.
Independent review.

Give it a goal. Fresh sessions work through the plan one step at a time, and an independent judge checks the result against your goal before anything is called done. When something falls short, the plan changes on the fly, and memory carries what each round learned into the next.

Agent loopLoad 3.4 sReady

CuttleCode on your phone

Step away.
Stay in control.

Everything you can control on desktop, you can control from your phone. Start sessions, manage projects, run workflows, and review your agents’ work.

No app store. Just open CuttleCode.

Add it to your home screen as a progressive web app. Easy setup. Secured via Tailscale

Your agents keep running on your connected host. Your phone puts you in control.

Intents

Write the intent.
Review the PR.

Describe your goal. Your agent researches the repo and writes a plan. Turn on autoplay to start work sessions automatically, and chat with the agent along the way to ask questions, give direction, or start another session. Everything stays together under one intent.

  1. 01Describe your goal

    Add rate limiting to the public API. Make it configurable and cover it with tests.

  2. 02Research and create a plan

    Researching

    • src/api/router.ts
    • src/api/middleware/auth.ts
    • src/config/flags.ts
    • src/lib/redis.ts
    • test/api/router.test.ts

    Plan created

    • Add limiter middleware (token bucket)
    • Wire the RATE_LIMIT flag into config
    • Mount on public routes only
    • Cover burst and steady state in tests

  3. 03Do the work

    Autoplay starts sessions to carry out the plan. Every session stays grouped under this intent.

    Add rate limiting to the public API1 session2 sessions

    • Build the limiter middleware

      Started by autoplay

    • Write burst and steady-state tests

      Started by autoplay

  4. 04Chat and orchestrate

    Ask for an update, give direction, or have the agent start another session—all from the intent’s chat.

    You

    How’s the rate limiting coming along?

    Cuttle

    The middleware and config flag are in place. The test session is covering burst traffic.

    You

    Start a separate session to review the changes.

    Cuttle

    I’ve started a review session under this intent. It’ll check the implementation and test coverage.

    Add rate limiting to the public API2 sessions3 sessions

    • Build the limiter middleware

      Started by autoplay

    • Write burst and steady-state tests

      Started by autoplay

    • Review API rate limiting

      Started from chat

  5. 05Review

    ccc/intent/rate-limit → main

    feat: rate-limit public API behind config flag

    +412 −87 · 9 files changed

Inside CuttleCode

01 / A shared starting point

Give every session the project context.

Explore Atlas Commerce: a storefront, payments API, and design system with shared instructions, a review skill, and a scoped agent.

Recorded in a dedicated example workspace. Silent walkthrough; each chapter has an accompanying description. Read the full walkthrough

Session wall

See the whole fleet.

Keep running sessions and conversations waiting for you together on one resizable wall.

Talk to your projects

Slack and Discord integration.

Policy tiers

  • unrestricted
  • wiki_only
  • chat_only
  • custom

Connect and authorize your bot first. Restricted chat policies are enforced by the Claude terminal runner.

01

DM → session

DM your bot, choose a project, and continue the conversation with an agent that has that project’s context.

02

Session → conversation

Ask an agent to post an update to your connected channel or thread. Follow up there and keep the work moving.

03

Channels

Provision a Discord project channel or bind an existing Slack channel. Start a thread and its session replies land where you asked.

04

Identities & policies

Every chat user maps to a CuttleCode identity with a policy tier that decides what they can touch.

And much more

This is the short version.

One page can't show everything CuttleCode can do. The rest lives one click away.

  • Context harness
  • Forge
  • Linear & webhooks
  • Worker nodes
  • Local inference
  • Computer use
  • Surfaces
  • Governance
Take the deep dive

Download the beta

Bring the idea. Take the helm.

Try the CuttleCode beta. Your agents, your hardware, your final say.

Download (Beta)Other platforms