Read the repo, write the rules.
Sign in with GitHub once and we read any repo you have access to — public or private. Or skip GitHub entirely and pick a local folder; the picker reads in your browser, only the curated slice you approve gets sent up.
We score every path for rules-relevance — manifests, configs, READMEs, entry points — and send the top ~12 (Free, Haiku 4.5) or ~24 (Pro, Sonnet 4.6) files to the model. The output is rooted in your actual conventions, not boilerplate.
Pro retunes the base markdown per tool — Cursor's MDC bundle with globs, AGENTS.md ordering, Windsurf voice, Copilot brevity. Each format reshaped for its agent, not just renamed.
The cost of starting cold.
Devs thought AI made them 20% faster. The clock said 19% slower. The gap is context.
AI saves the average dev 3.6 hours a week. Fifteen minutes of re-priming per session erases all of it.
90% of developers lose 6+ hours a week to context-switching. Your AI agent is not exempt.