Problem
memcp has two backends (mem0, in-memory) but hasn't validated the Protocol against a fundamentally different architecture. Cognee is dataset-oriented with a knowledge graph pipeline — a strong test of backend agnosticism.
Proposed Solution / Alternatives
Implement CogneeBackend(MemoryBackend) mapping Cognee's remember/recall/forget/update endpoints to the memcp Protocol. Includes tenant-to-dataset mapping and any shims needed for Cognee-specific behavior (inseparable from adapter work).
Key mapping decisions: user_id/scope → dataset/tenant model, content storage via remember endpoint, search via recall with appropriate SearchType routing.
Problem
memcp has two backends (mem0, in-memory) but hasn't validated the Protocol against a fundamentally different architecture. Cognee is dataset-oriented with a knowledge graph pipeline — a strong test of backend agnosticism.
Proposed Solution / Alternatives
Implement
CogneeBackend(MemoryBackend)mapping Cognee's remember/recall/forget/update endpoints to the memcp Protocol. Includes tenant-to-dataset mapping and any shims needed for Cognee-specific behavior (inseparable from adapter work).Key mapping decisions: user_id/scope → dataset/tenant model, content storage via remember endpoint, search via recall with appropriate SearchType routing.