Give your agents context they can trace.
A self-hosted semantic layer that connects source evidence, agent drafts, and human review—then publishes context your agents can use.
Open source · Apache 2.0 · Public alpha

From source to shared context.
Build with your agents. Keep review and publication in human hands.
-
Connect sources
Scoped metadata and cited document evidence.
-
Agents draft
Definitions, joins, and metrics with evidence.
-
Humans review
Accept supported claims. Keep uncertainty visible.
-
Publish context
Share reviewed, current knowledge through Model Context Protocol (MCP).
Models run in your chosen agent harness. Your separately authorized source tools execute queries and verify results. Read the build and review flow.
Make meaning explicit.
A shared knowledge graph connects definitions, source bindings, evidence, and review state—so meaning stays inspectable.
Explore the knowledge graph- Typed contracts
- Define grain, joins, and metrics with exact evidence bindings.
- Open graph standards
- Resource Description Framework (RDF) exports use Simple Knowledge Organization System (SKOS) vocabulary and PROV Ontology (PROV-O) provenance.
- Optional validation
- Shapes Constraint Language (SHACL) checks graph structure. Evidence and human review still matter.
Get started locally.
Docker Engine with Compose and Python 3.11+ on Linux or macOS.
git clone https://github.com/Quiet-Victory-Labs/knowledge-layer.git
cd knowledge-layer
./knowledge-layer init --examples
./knowledge-layer start
Open 127.0.0.1:4173, use the launcher’s private bootstrap token to create your local owner account, and follow Setup.