Context with a home.
Keep project instructions, repo references, skills, and agents together. Reuse what matters across sessions without rebuilding the briefing every time.
Deep dive
Everything beyond the front page: how context is composed, how skills improve, what starts work, where it runs, and what stays on the record.
Context is the foundation
The frontend, the API, the shared library. Your agents need to understand how they fit together. CuttleCode gives that understanding a place to live.
An example structure · inspect the real one in the tour
Keep project instructions, repo references, skills, and agents together. Reuse what matters across sessions without rebuilding the briefing every time.
The context harness shows the composed instructions, context layers, token estimates, and authoring findings. Inspect the starting point before asking an agent to work.
Start at the project to reason across repositories, or work in a specific repo. Give specialized agents the skills they need for their part of the job.
Inside your workspace
Keep sessions, agent loops, and their progress together. Open a capture for a closer look.
Sessions
A Claude Code session improving search in an example project, with real tool calls, a test summary, and model context in view.
Real captures from an example CuttleCode workspace · select one for a closer look
Forge
Turn the instructions your agents use into versions you can compare. Forge drafts alternatives, evaluates them against reserved cases, and brings eligible improvements back for your review.
Skills, agent prompts, and Ralph strategies can all enter Forge. Scheduled evolution proposes improvements; automatic promotion is a separate opt-in for artifacts you mark low-risk.
The current version—the champion—keeps serving your sessions while Forge works on alternatives.
Work can start anywhere
Start with the tools your team already uses. Bring the request into a project with the context and permissions to act on it.
Bind a project to Linear, choose the assignees to follow, and adopt matching stories as intents. Autoplay carries one from research toward a PR, with progress and comments tied to the original story.
Run a recurring review on a cron schedule, or start work from an incoming webhook. Either can trigger a session or a pipeline. Webhooks need a rotatable bearer secret — the URL alone is not enough.
Expose a project’s HTTP service to build an internal agent chat. Map guest and signed-in users to gated actions, load the context each role needs, and enforce tool permissions at the agent boundary. You configure it, with your own authentication and authorization.
Multiple workspaces
Keep each client, company, or personal project in its own workspace, with separate credentials, members, and settings.
Client workAtlas Studio
Your next ideaAfter Hours
The details add up
Quiet features you end up using every day.
Projects, members, Git credentials and model defaults live per workspace.
Unlimited devices on every cloud plan. Invite and manage members.
Choose what syncs per machine. Transcripts stay on the instance.
Knowledge & platform
Surfaces
Manage work from desktop or phone. Start sessions, follow progress, and keep your agent choices and account preferences close at hand.
Built-in Tailscale support takes your home-hosted instance beyond your home network. Set up Tailscale on your host and phone, or host CuttleCode online on your own VPS. You run the instance and manage its connection.
browser
Pair the Chrome extension, explicitly share a tab, and let a human-started session work with the page in your signed-in browser. Stop sharing when you’re done.
Requires extension setup and browser support enabled in Cuttle. Only shared tabs and granted sites are reachable; scheduled automation cannot drive this browser lane.workspaces
Keep personal and shared workspaces separate. Link your machines, invite collaborators to shared workspaces, and manage projects and devices from your account. Cloud sync can be controlled per workspace on each machine.
Shared workspaces and cloud limits depend on your plan. Cloud run history carries status, verdicts, and cost summaries; session transcripts and terminal output stay on the instance.Governance
Control who can act, give each session the right context, and keep the evidence close to the work.
Set roles and permissions for your workspace. Keep credentials in encrypted storage.
Project instructions and scoped skills follow the work across supported agent harnesses.
Use isolated worktrees so concurrent changes stay separate until you review them.
Inspect audit events, decision records, and token usage alongside the work they describe.
Download the beta
Try the CuttleCode beta. Your agents, your hardware, your final say.