Developers

The technical contract

For the engineer whose team wants to use this. Sixty seconds.

What it is

TreeGates governs how AI-assisted work moves from ticket to merge. Your team keeps working with AI the way it already does. You describe the project and connect your stack. TreeGates builds the workflow from that: the graph, the worker nodes with their rules and prompts, the testing and QA gates. You sit in the loop wherever you decide. You govern your own AI, and the outcome is yours.

What runs where

• Your code, credentials, and AI-provider keys never leave your environment.
• TreeGates itself runs on your machines. The UI opens in your browser, served from localhost.
• Our cloud holds accounts, seats, billing, and workflow templates. It never needs your source code or keys.

YOUR ENVIRONMENT                        TREEGATES CLOUD

┌─────────────────────────────┐
│ ┌─────────────────────────┐ │         ┌─────────────────────┐
│ │ you                     │ │         │ accounts and seats  │
│ │ terminal · browser UI   │ │ ◄─────► │ billing             │
│ └──────┬─────────▲────────┘ │         │ workflow templates  │
│   work │         │ decisions│         └─────────────────────┘
│ ┌──────▼─────────┴────────┐ │
│ │ TreeGates engine        │ │         what crosses: seats and
│ │ gates · loops · records │ │         templates. never your
│ └──────┬─────────▲────────┘ │         code, never your keys.
│dispatch│         │ evidence │
│ ┌──────▼─────────┴────────┐ │
│ │ AI workers ► your repo  │ │
│ └─────────────────────────┘ │
└─────────────────────────────┘

What it can do

Each AI worker gets a role you define and permissions you grant one by one: read, write, review, push, merge. An AI that can write code still cannot merge it unless you said so.

What happens on failure

• Gates fail closed. If the evidence is missing, stale, or failing, the work stops moving.
• An approval binds to the exact commit that was reviewed. Old approvals cannot authorize new code.
• Retry loops are bounded. Three fix rounds by default, then the work stops and a named human is asked.

What it learns

Every run ends in a retrospective. Lessons go into a knowledge base your team owns, each cited back to the run that produced it. Later runs check it before repeating a mistake, and repeated failures become suggested changes to your workflow.