FAQ

Does Kairn upload my code?

No, not by default. Core repository indexing, ranking, and session state run locally. Optional aggregate telemetry is off until you enable it.

Does Kairn require an account or API key?

No Kairn account or API key is required for the local product. The coding agent you connect may have its own provider requirements.

Does reducing context make quality worse?

It can if done carelessly, which is why Kairn does not treat token reduction as a standalone success metric.

Public token comparisons count only when both attempts pass their quality and required-source gates. Source recall, excluded files, verification, and diff scope take precedence over a smaller token total.

Are the token numbers estimates?

The benchmark comparisons shown on the evidence page use observed model-token counts on selected pass/pass rows. They are labeled historical because they predate the calibrated-ranker protocol.

Aggregate beta telemetry may estimate context suppression for product diagnostics, but the site does not present that estimate as observed billing usage.

Does Kairn always send context?

No. Kairn can send source, send a path hint, request missing evidence, or stay silent. Silence is correct when added context is unlikely to help.

Is Kairn just RAG for code?

Retrieval is part of Kairn. It finds relevant code locally, then keeps track of what the agent has already seen, what changed, and whether a reported failure offers a new clue. It may send source, give a shorter hint, or send nothing rather than repeating unchanged code.

Other retrieval systems can make similar choices. Kairn has not published a head-to-head result against a separate RAG product; the historical results compare Kairn-assisted coding agents with unassisted runs.

Is Kairn generic agent memory?

No. Kairn is deliberately scoped to precise, local evidence from the current repository checkout and the active coding session.

How is Kairn different from ReasonBlocks?

Kairn focuses on local current-repository evidence for coding agents it does not control. ReasonBlocks focuses on agent-trajectory steering and reasoning reuse through supported programmable agent loops and hosted services.

ReasonBlocks is not described here as closed source, expensive, or universally difficult to configure. Its Python SDK is MIT-licensed. Its full steering architecture uses hosted services and has a separate trust disclosure. The planned comparison tests Vanilla, ReasonBlocks, Kairn, and the combination under the same model, tools, limits, and verification contract; no comparative outcome is claimed before those runs pass protocol checks.

Which agents are validated?

Codex is the most-tested beta path. Local MCP delivery has benchmark evidence. Claude Code, Gemini CLI, VS Code/Copilot, and other clients may have supported setup paths without equivalent published efficacy validation yet.

Is Kairn open source?

The published v0.2 beta package is Apache-2.0 licensed. The GitHub source repository is still private, so Kairn should not yet be described as a publicly available open-source repository. A public source release is planned for v0.3.

Is this a stable release?

No. usekairn@beta is publicly available at 0.2.0-beta.2, but it is a beta rather than a stable release. The stable v0.3 distribution and public GitHub source release are still planned.