About 2,000 tokens.
The ticket's title, description and plan pick the docs it reads. Fewer tokens, a faster start.
You don't manage your project context. Fredrin does.
The Self-learning Brain gives every Worker what its ticket needs, and keeps what it learned.
Search that doesn't read everything. A slice per ticket, not the whole repo.
Local and offline.
Full-text search runs in the worktree. No vector database.
87% recall.
Of the live docs a decision cites, on Fredrin's own 2,207-doc repository.
It learns as you ship. Recorded as your agents work.
Workers write back.
What a Worker learns ships in its own pull request.
Decisions get recorded.
A gate at finish decides which changes deserve a decision record.
Settled stays settled.
Past decisions show up as a heads-up when a build starts.
Your project's own language. Built up ticket by ticket.
Plain markdown, in your repo.
A glossary, concept docs, traps and decision records, versioned in git.
Your terms, not synonyms.
Glossary terms ride in each ticket's slice. With Claude Code, a matching trap shows before an edit.
It matches words, not meaning.
A doc with different words and no links can be missed.