What is getff

getff is two layers around one idea: conventions your AI agents can't silently bypass. The tagline on our landing page says it in one line: «AI DX for your codebase: conventions AI agents can't silently bypass — and an AI-run dev environment around them.» Each half of that sentence is one layer, and each layer is at a different maturity — labeled honestly below.

Rules from live docs — beta

The first layer compiles your conventions into the toolchain gates your project already runs: ESLint rules and husky hooks for npm projects, gate generation for Rust. A convention that lives as a rule file fails loudly at edit time, pre-commit, pre-push or CI — not politely in prose the agent parsed and ignored. Today this layer covers TypeScript/JS, Rust, Python and Go stacks, with more toolchains on the roadmap. And the flagship honest limit: an executable AGENTS.md today means getff's own repo — generating yours from your conventions is the next milestone, not a shipped feature.

The rules layer ships as a beta.

The AI factory — experimental

The second layer is the AI-run half: a multi-model dispatch pipeline where a kickoff goes in and a harvested branch comes out. Tasks are tiered by a fixed rubric — who plans, who implements, who reviews — so cheap mechanical work runs on a cheaper model and design-heavy work plans on a stronger one. The factory installs on top of the rules layer: the factory profile ships the same rule-proving steps, so dispatched work happens in a repo where those gates are installed and proven.

The factory layer is experimental: it expects a specific operator runtime and its capability surface degrades in named ways when pieces are missing — see Degradations.

Start where you like

• Living with the beta layer day to day: Daily cycle — rules.
• What the experimental factory actually is: Overview — multi-model pipeline.