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.

Is the store format proprietary? No — KAUT implements the vendor-neutral Open Knowledge Format (OKF) v0.2: typed-markdown concept documents that meet OKF's conformance bar in place (kaut okf check audits, kaut okf stamp backfills legacy docs), and kaut okf export projects a store into a fully idiomatic OKF v0.2 bundle — provenance (sources), trust (generated/verified, with owner-gate approvals surfacing as the human-reviewed tier) and index/log conventions included. KAUT's freshness/trust fields ride as OKF-protected extensions.

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 tech stacks does it work with? Bootstrap and the whole knowledge loop are stack-agnostic — any git repository works. The mechanical map layer has stack adapters with auto-detection: Vue, Java/Kotlin + Spring (Gradle/Maven), Next.js, Express/Nest, FastAPI/Flask, PHP (Laravel/Symfony), Flyway-style SQL migrations, docker-compose. No recognizable stack = the map layer simply stays empty; everything else works the same. See Connecting Your Project.

Which operating systems? Developed and tested on macOS and Linux (CI runs Linux, Node 20 and 24). Windows is not supported.

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.