Skip to content

Memory and GraphRAG

FabrCore 2.0 · Release and package availability

These guides track the current 2.0 source. Stable 2.0.0 publication is pending; package commands show the release target. Until it is published, follow the source quick start or use a matching available prerelease set. Release migration · Runtime modes

Give agents durable knowledge

Memory keeps scoped knowledge across conversations. GraphRAG ingests and searches shared documents using vector and graph data. Both implementations now ship in FabrCore.Host.

Enable SQL mode and configure default chat plus 1536-dimensional embeddings. No separate Memory or GraphRAG package or manual service registration is needed. Contracts live in Core; remote Memory clients and harness integration live in SDK. Namespaces remain compatible where applicable.

Memory is an explicit agent choice

Use IAgentMemoryProvider to obtain a scope-bound IAgentMemoryService, or select the agent-memory plugin. Derive scopes from trusted application context. WithMemory and WithMemoryLifecycle provide explicit harness integration; SQL mode does not attach memory automatically.

Default recall selects headers and loads primary chunks. The hot index defaults to 20 entries and a 3,000-token budget; warm selection chooses five memories from up to 200 headers. Automatic consolidation and summary trees stay off. Semantic/hybrid retrieval and other experimental paths are opt-in.

GraphRAG separates source from extraction

Ingest through IKnowledgeIngestionService with KnowledgeIngestionRequest. Search through IKnowledgeSearchService with trusted allowed scopes. Lossless source sections are separate from overlapping vector chunks, with bounded extraction batches, concurrency and malformed-response retries.

Optional schema output, evidence quotes, source provenance and caches support different workloads. Evaluate quality with your corpus and selected model rather than assuming a universal speed or accuracy improvement. Database startup applies the registered GraphRAG migrations.

Use the Memory skill and GraphRAG skill for code examples and detailed API guidance.