CogniGraph / Source
Read the code.
Follow the decisions.
CogniGraph is built in Rust by Gedank Rayze, LDA, in Lisbon. Its source workspace contains the database, query engine, HTTP server, tools and Enterprise components, alongside the contracts and evidence used to review them.
01 / Where the behavior lives
Twelve crates. One product.
The source repository and engineering documentation are public on GitHub. The Rust workspace separates storage, query planning, server integration, authentication, embeddings and Lua. Both Community and Enterprise use CogniGraph's own Native storage engine.
01 / Community
Store records. Query their connections.
The document model, query engine and durable Native store share an explicit backend contract.
cognigraph-coreDocument and graph types, backend contracts and errors.cognigraph-queryCGQL grammar, validation, planning and execution.cognigraph-nativeNative storage, traversal, text and vector indexes.
02 / Community
Program the steps around a query.
Lua connects database operations with application-specific logic, under execution budgets and explicit permissions.
cognigraph-luaSandboxed runtime, graph bindings and execution controls.
03 / Community
Find meaning. Reuse useful work.
Embedding providers and query caches support retrieval alongside the database's search indexes.
cognigraph-embeddingsA shared provider interface for local and remote embeddings.cognigraph-cacheQuery normalization, semantic matching and memory/persistent cache implementations.
04 / Community foundation
Connect an application. Operate the service.
The HTTP server, authentication and administration tools connect the engine to its users. Enterprise builds add governed and tenant operations to the server and CLI.
cognigraph-serverHTTP endpoints, request handling and service integration.cognigraph-authUsers, tokens and role-based access control.cognigraph-cliAdministration, operations and edition-specific tooling.
05 / Enterprise
Construct knowledge with an evidence trail.
Construction rules, signed authority and verified artifacts support the path from proposed relationships to governed releases.
cognigraph-constructSemantic Neurons, evidence grounding and recall/restraint evaluation.cognigraph-governanceCanonical records, signatures and portable verification primitives.cognigraph-artifactsContent-addressed artifact verification and offline recovery primitives.
Alongside the Rust workspace
| Path | What to look for |
|---|---|
ui/ | The React management console. |
docs/ | Runtime contracts, operations, decisions, issues and change records. |
fixtures/ | Examples, reproducible experiments and sealed result packages. |
Read the database overview before choosing a subsystem, or start with the developer introduction for the HTTP boundary.
02 / Source & permissions
Available under stated terms.
CogniGraph uses source-available licensing. Community components use the Functional Source License with an Apache 2.0 grant two years after each release. Enterprise components have separate commercial terms. Source access and permission to run a component in production are different questions.
The edition and licensing overview explains the practical boundary. The license texts and any executed commercial agreement determine the actual rights.
03 / Make a useful change
Bring a reproduction.
A clear report identifies the component, the data shape, the expected result and what happened instead. Use a small synthetic reproduction when the original records contain private information.
Changes are reviewed against the repository's agent instructions and checks. Contributions require the Contributor License Agreement, signalled by a Signed-off-by: trailer on every commit. Read the agreement before contributing.
For enquiries about the project, visit the Contact page.
04 / Claims you can examine
A result needs its context.
Engineering evidence lives with the code. Experiment records identify the corpus, model, policy and scoring units that bound a result. Sealed fixtures preserve the bytes behind a measurement; later measurements belong in new records.
The issue registry retains resolution evidence, decisions explain contract changes, and the implementation plan separates delivered behavior from the next work. These records give a technical review somewhere concrete to begin.
Read the limits with the results.
Every experiment record states its corpus, model and scoring unit, so a number carries its own boundary. Roadmap items are labelled as roadmap even where the code already holds their foundations.
Plan an evaluation around your question