Provides the semantic memory and context foundation for the Flow system, and defines how models and tools are integrated and invoked verifiably.
- KG: Manages entities, relationships, context, causality, and provenance as a distributed graph.
- MCP: Standardizes the description, requirements, invocation, and verification of external models, tools, and functions.
- KG: Provides semantic context and long-term memory for agents.
- MCP: Handles the provenance, invocation, and verification of models and tools used by agents and DAG tasks.
- Data Structure: IPLD-compatible object graph.
- Backend: CRDT-backed persistence (e.g., using Any-Sync) for decentralized consistency.
- Nodes: Represent entities (users, agents, tasks, data, concepts), events, etc. Identified by CIDs or DIDs.
- Edges: Represent semantic or causal links between nodes.
- Layered Model: Conceptually layered for different types of information:
- Entity Layer
- Context Layer
- Semantic Layer (Schemas, Ontologies)
- Causal Layer (Execution Traces)
- Provenance Layer (Origin, Signatures, Proofs)
- Schemas: Supports JSON-LD, RDFS/OWL-lite for defining object types and relationships.
- Access Control: VC-based permissions govern read/write access to graph partitions.
- Links KG nodes to:
- Agent SLRPA phases (providing context for Sense, Reason, Learn).
- DAG task inputs/outputs.
- Agent-to-Agent messages.
- Supports scoping context by:
- Time
- Logical relevance
- Privacy constraints (VCs)
- Enables runtime resolution of context needed by tasks/models.
- Tracks provenance of context used.