FAQ

FAQ

Do I need TAUT (or any orchestrator) to use KAUT? No. KAUT is a standalone product: clone the engine, connect the MCP server to your harness (or call the CLI from any skill or prompt), done. TAUT integrates KAUT natively if you use it — a composition, not a dependency.

Is this just another agent memory system? No. Memory remembers conversations — episodic, personal, unverifiable. KAUT is the project's documentation, AI-first: source-bound, commit-anchored, freshness-checked, readable by agents and humans alike. See Why KAUT.

Is this RAG? No. No embeddings, no chunking, no ranking. A small set of distilled docs read whole, each with provenance and a git-computed freshness verdict — storing deliberately only what is not cheaply derivable from code. RAG and KAUT answer different questions and coexist fine.

Is this an auto-generated wiki? No. Nothing enters as unverified generated prose: every doc needs typed sources and a commit anchor; mechanical layers are regenerated, not hallucinated; judgment-tier knowledge passes a human gate. And a KAUT doc cannot rot silently — its sources are diffed on every read.

Will it commit anything into my repository? No. At most one git-ignored pointer file. The knowledge store lives outside the repo.

Where does the data live? By default under ~/.kaut. To keep knowledge in a dedicated folder with its own lifecycle, record a redirect once: kaut home <dir>. Every later caller — CLI or MCP — resolves the location by itself; nothing to export.

How do I back up the knowledge base? kaut backup — the whole data home (stores with their git history, registry, setup record) becomes a dated, versioned .tar.gz under <data>/backups/, with zero external dependencies (readable by any tar tool too). kaut restore latest brings it back and never overwrites existing data without --force.

What does it cost to run? Freshness checks are pure git comparisons — no model calls, no index to rebuild, no daemon. The only "expensive" step is the first map build (seconds to minutes).

What if a stored fact is wrong? Every fact carries provenance and a trust label; agents treat low-trust and coarse-altitude answers skeptically and verify against code (the standing discipline: knowledge informs, it never authorizes). The store keeps full git history, so any entry can be traced and rolled back; the stale-misled outcome tag makes failures visible in kaut digest.

Does my team have to adopt it? No. Local-first: one developer installs it and benefits. If the team adopts it later, the store is a normal git repository — reviewable, diffable, portable.

Which models/harnesses does it work with? Any MCP-capable harness via the bundled server, and anything that can run a CLI. The engine is harness-neutral by design.

Can I run it in CI? Yes — kaut doctor and kaut stale --json are non-interactive, exit-code honest, and take no lock on the read path.

Why "Actualization Under Trust"? The two mechanisms the name promises: actualization — knowledge is kept current structurally (anchored sources, verdicts, the touched/refresh loop), and under trust — every answer carries explicit trust routing, and every write passes a gate that keeps humans in charge of judgment-tier knowledge.