Forge with DeepLake and HiveMind
From Knowledge Graph to Living Memory
How Forge combines ActiveLoop open-source memory layers with governed graph intelligence to create a stronger foundation for human and agent memory.
132
Benchmark queries
100%
Overall hit@5
72
Recovered targets
Graph authority
Local recall
Governed results
Why the combination matters
Institutional memory needs powerful recall, but it also needs trust. Retrieval should find relevant context even when the exact wording is unknown, while still respecting provenance, lifecycle state, archive rules, and governance boundaries.
Forge uses Neo4j as the authoritative knowledge graph, DeepLake as a local vector recall layer, and HiveMind as a local memory continuity layer. Sidecars suggest candidates, then the graph governs what can be returned.
A fully covered active memory set
Forge currently tracks 8,877 capsules, including 2,207 active capsules in the live retrieval set. Those active capsules are represented in both local memory sidecars.
That full active coverage gives Forge a steadier retrieval baseline across graph search, vector search, and memory search without making an accelerator store the primary truth.
Operational principle
DeepLake helps Forge ask which records are semantically close. HiveMind helps preserve local memory around capsule lifecycle. Neo4j answers what the record is, how it is connected, and whether it is valid to use.
This keeps recall expansive while keeping the final result anchored to governed source records.