Knolo MCP server

Knolo

The verifiable knowledge layer for trustworthy AI

Knolo turns organizational knowledge into portable, deterministic, and cryptographically verifiable runtime artifacts.

AI products are becoming easier to build. The hard problem is making their knowledge dependable: traceable to evidence, reproducible across environments, safe to move, and defensible under review. Knolo is infrastructure for that problem.

Knowledge should be an asset you can inspect, verify, migrate, and run—not opaque context trapped inside an application.

V6 · KAR · V5 Knowledge Images · V4 compatible · Local-first · Deterministic · Verifiable

Current release

This tree is Knolo 6.1.0. V6 adds KAR: deterministic retrieval of a minimum evidence set from a V5 Knowledge Image and a Committed Evidence Graph. 6.1.0 adds verifiable composition: offline citations of images, packs, graphs, and bounded excerpts, plus attestations. Lexical retrieval, V5 images, and V4 packs stay in place.

import {
  createKarSession,
  evaluateKar,
  verifyKar,
  explainKarResult,
  createKarDossier,
  createKarCoverage,
} from '@knolo/core/kar';

The root @knolo/core entry does not load KAR. @knolo/core/experimental/kar remains a compatibility entry for this cycle. SATISFIED means the declared evidence requirements were met. It does not mean a claim is proven. UNSATISFIEDEVIDENCEREQUIREMENTS means those requirements were not met. It does not mean the claim is false.

The npm packages in this repository are 6.1.0. @knolo/cli depends on @knolo/core ^6.1.0. The other packages that depend on @knolo/core stay at ^6.0.0. Publishing that version, and moving the npm latest dist-tag, is a separate step. @knolo/reflex stays 0.2.0. @knolo/evidence-gate stays 0.1.0. Rust, Python, and the ICP canister stay 5.5.1 and do not implement KAR or composition. The bundled contracts, policy, and operations Domain Packs are review-required. The local tag v6.0.0 stays on the 6.0.0 commit.

Start with the KAR guide, the composition guide, the 6.1.0 release notes, examples/kar-v6, examples/kar-coverage, and examples/composition-v1.

The opportunity

The next generation of AI products will compete on the quality and trustworthiness of their knowledge layer. Models may change quickly; durable value accumulates in the evidence, policies, retrieval behavior, runtime state, and operational history surrounding them.

Knolo provides a neutral substrate for that layer:

• Portable knowledge — package a knowledge base as a self-contained Knowledge Image rather than a proprietary database export.
• Verifiable state — every V5 image has deterministic encodings, domain-separated SHA-256 roots, bounded sections, and a commit history.
• Traceable answers — preserve source, chunk, claim, citation, and evidence identity through retrieval and agent execution.
• Reproducible behavior — use deterministic query plans, receipts, migration records, and runtime diagnostics to make results inspectable.
• Composable infrastructure — keep storage, verification, retrieval, policy, and agent execution separate so products can adopt only what they need.

This is the strategic thesis behind Knolo: a trusted knowledge runtime can become shared infrastructure beneath many AI applications, agents, and enterprise workflows.

Why Knolo

Shortened. The full README is on GitHub.