How-to
How to start a project with CatWrangler
Connect CatWrangler, point your agent at it, and describe what you want to build in plain language. It hands your agent the relevant code and decisions, gives it its own line of work, validates every change, and ships on merge. No version control to learn.
Starting a project with CatWrangler is mostly a matter of saying what you want. You connect once, point an agent at it, and describe the work — the system handles the rest, server-side, where you never have to see it.
What you'll need before you start
Not much. CatWrangler is a hosted service on your own isolated server, so there's nothing to install or self-host. You bring an AI agent and an idea — or an existing repo you want to keep building on.
- A new project, or a repo to bring in (you can take it back out in one shot — no lock-in).
- An AI agent you already work with. It connects to CatWrangler and works through it.
- No version control knowledge. You and your agents never see Git, a branch, or a merge.
What happens once you're connected
From the first connect, CatWrangler reads your existing code and docs so nobody starts from a blank slate. Your agent describes its intent and gets a pre-digested briefing — project, structure, the decisions that matter, and where to work — instead of crawling the codebase. The write path is the only gated part; reading and exploring stay open.
- Describe a change in plain language and get the exact files and functions, plus the reasoning behind them.
- Every change records its decision — the choice, the reasoning, the alternatives ruled out — bound to the code, so the 'why' survives.
- On submit, a change is validated: it must cover the decision it claims and must build and pass tests before it can merge.
- 01
Connect to CatWrangler
Point your AI agent at your CatWrangler server. It's a hosted service on an isolated server of your own, so there's nothing to install and nothing to self-host. On first connect, CatWrangler reads your existing code and docs — including large or legacy codebases — so you're never starting from a blank slate.
- 02
Let your agent get its bearings
Instead of crawling the repo, your agent describes what it's trying to do and gets a pre-digested briefing: the project, its structure, the decisions that matter, and where to work. Before it builds, the system surfaces code and decisions that already do the job — so it extends what's there rather than duplicating it.
- 03
Describe what you want to build
Say it in plain language. CatWrangler hands back the exact files and functions to touch and the decisions behind them, so your agent isn't guessing or grepping. Incompatible choices are surfaced and negotiated here, before any code exists, while they're still cheap to change.
- 04
Let the gates and decisions do the work
Your agent automatically gets its own line of work — the shared trunk is never written directly. Every change records the decision behind it. On submit, the change is validated: it has to cover the decision it claims and build and pass tests before it can go anywhere.
- 05
Merge to ship
On merge, an AI combines the compatible changes and routes anything it can't safely resolve back to the responsible agent, blocking until it's clean — work is never silently overwritten. Then your change builds, validates, and deploys itself, with no separate pipeline. Deploys swap in with no downtime, a bad one rolls back, and a tagged prior state can be restored.
“I asked which files needed refactoring and where the seams were — and got back a structured briefing with citations, the constraints that applied, and the next moves. One call set up the whole task.”
Related questions
Do I need to know Git to start a project?
No. Real version control runs underneath, but you and your agents never see Git, a branch, or a merge. The target isn't learning version control — it's being more effective with your agents. CatWrangler handles all of it server-side.
Can I start with an existing codebase instead of from scratch?
Yes. Bring a repo in and CatWrangler reads your existing code and docs on first connect, so your agent gets a real briefing instead of a blank slate. It works on large and legacy codebases — and you can take your repo back out in one shot, with no lock-in.
Keep reading
Vibe-Engineering
Many agents. One codebase. Zero collisions.
Point your agents at CatWrangler and build — the discipline runs underneath.