Native gRPC channel
Connect-RPC compatible protocol over Unix socket or TCP with HMAC nonce challenge-response authentication. Local-first by default, TLS available for remote deployments.
Hermes support
Native Python memory provider for Hermes Agent.
hermes-memory-libravdb connects Hermes Agent to libravdbd as a native memory provider and context engine. Replaces the default lightweight memory with durable, daemon-owned vector storage — scoped across session, user, and global namespaces.
Hybrid retrieval
Semantic similarity, scope affinity, recency weighting, and summary quality combine into a single ranked recall pass — not a chain of separate lookups.
Continuity-aware assembly
Recalled memories are fit into a bounded prompt budget while preserving the most recent context. Agents stay coherent without overflowing the context window.
Background turn sync
User and assistant turns are queued for daemon ingestion asynchronously. The Hermes UI path never blocks on a memory write.
Local-first inference
Embedding, compaction, and retrieval run on local hardware by default. No cloud API calls, no token billing, no data leaving the machine.
Full CLI surface
Health checks, semantic search, namespace flush, NDJSON export, journal inspection, and index rebuild — all available through hermes libravdb subcommands.
60+ configuration keys
Fine-tune embedding profiles, scope weights, compaction schedules, decay curves, and retrieval budgets through the provider config file at $HERMES_HOME/libravdb.json.
Hermes lifecycle integration
Memory writes, delegation results, pre-compression hooks, session end, status, export, flush, and journal commands stay wired into the native Hermes tool surface.
pip install hermes-memory-libravdbPython >= 3.11 libravdbd daemon gRPC + HMAC auth Unix socket or TCP 60+ config keys pip / pipx install
GitHub — hermes-memory-libravdb · Talk about Hermes deployment