FAQ
Short answers for public beta testers.
Does Kairn upload my code?
No, not by default.
Does Kairn always speak?
No. Suppression is part of the product.
Does reducing context make quality worse?
That is the main risk Kairn is built around.
Kairn does not try to save tokens by blindly cutting context. It saves tokens by removing repeated or low-value context, keeping known source state across turns, and staying quiet when extra context is likely to add noise.
The public evidence only counts clean savings when the task still passes. Rows where Kairn improves quality are reported separately.
Is Kairn just RAG for code?
No. RAG retrieves candidates. Kairn governs execution context: what reaches the agent now, what stays out, and what evidence is needed before adding more tokens.
It can route source, shrink the prompt, stay quiet, or ask for one missing signal. That is the difference: Kairn governs context instead of blindly appending retrieval results.
Does Kairn replace RAG?
No. Kairn can use retrieval; it governs context after and around retrieval.
RAG finds relevant candidates. Kairn decides what should actually reach the agent during the task.
Is Codex required?
No. Codex CLI is the most-tested path, but MCP exists for editor and extension workflows.
Is this a public release?
It is a public beta and still a technical preview. Install with npm install -g usekairn@beta.