Bilinc is hosted infrastructure for AI agent memory. It provides verifiable commit, recall, audit, and recovery — not just vector similarity search.
Bilinc gives autonomous agents durable, typed memory with verification. Every commit is checked for consistency. Conflicting beliefs are resolved through AGM revision. Agents get audit trails, rollback, and recovery.
Bilinc Cloud is hosted infrastructure for AI agent memory. It provides commit, recall, audit, and recovery through a Python SDK, CLI, and MCP adapter. Unlike simple vector databases, Bilinc models agent state with typed memories, verification, and belief revision.
Bilinc stores memories as typed entries (semantic, episodic, procedural, spatial, working) with importance weights. When a new memory contradicts an existing one, the AGM belief revision engine resolves the conflict. Every commit can be verified through Z3 SMT proofs for provably consistent agent state.
Yes. Bilinc is source-available under BUSL 1.1 with an Apache 2.0 roadmap. The source is on GitHub at github.com/ReARCLabs/Bilinc. The public PyPI package bilinc 2.1.4 provides the cloud-only SDK, CLI, and MCP adapter.
Bilinc scores 98.0% R@5 and 0.933 NDCG@5 on the cleaned LongMemEval-s 500-question fixture with no LLM reranker or paid API. ConvoMem and LoCoMo numbers are internal ReARC regression checks, not official dataset claims.
Install from PyPI: pip install -U bilinc. Run bilinc start to initialize, then bilinc login --api-key <key> to connect to Cloud. The package currently provides cloud-only access through CloudClient, CLI, and MCP adapter.
Vector databases retrieve similar text. Bilinc is an agent memory/state plane: it preserves operational state with typed commits, provenance tracking, audit trails, contradiction detection, and recovery. It uses a vector index internally for retrieval but is designed for agent workflow infrastructure.
