Context Lattice (memMCP) gives agents reusable memory with HTTP MCP, federated retrieval, durable fanout, and storage guardrails so quality and costs remain stable under load.
One intelligent memory layer for apps and agents.
Replace repetitive long-context stuffing with reusable memory retrieval and sink durability.
Ships with Qwen by default and can plug into your preferred stack via Ollama, LLVM/llama.cpp, LM Studio, and compatible local gateways.
Write path is tuned for ~100 messages/sec with queue backpressure, fanout coalescing, admission control, retry workers, and retention sweeps to keep sinks stable under burst load.
Use gmake quickstart for first-run setup, secure bootstrap, and health verification.
gmake quickstartORCH_KEY="$(awk -F= '/^MEMMCP_ORCHESTRATOR_API_KEY=/{print substr($0,index($0,"=")+1)}' .env)"
curl -fsS http://127.0.0.1:8075/health | jq
curl -fsS -H "x-api-key: ${ORCH_KEY}" http://127.0.0.1:8075/status | jq '.service,.sinks'Secure mode is on by default. Keep using /health without auth, and include x-api-key for protected endpoints like /status, /memory/*, and /telemetry/*.
The orchestrator uses a learning schema from feedback signals to rerank results and improve retrieval precision over time. This is reinforced by RAG through Letta archival memory, alongside Qdrant, Mongo raw, MindsDB, and memory-bank fallback.
Read the detailed rollout notes on the Updates page.
Every write enters through the orchestrator, which records durable raw data, fans out to specialized stores, and continuously protects queue and storage health. Every search comes back through the same orchestrator so results can be fused, reranked, and improved over time from feedback.
Single orchestrator spine, explicit method stages, and parallel fanout branches to all write sinks.
Federated sources converge to the orchestrator spine, then reranked results return with learning feedback.
Benefit: one control plane for writes, retrieval, and policy.
Why: central coordination is what allows multi-source ranking and learning to compound.
Benefit: canonical project/file context store.
Why: keeps user-facing memory deterministic and compatible with MCP-native clients.
Benefit: high-speed semantic recall.
Why: vector retrieval gives broad relevance quickly before deeper reranking.
Benefit: durable source-of-truth write ledger.
Why: protects recoverability and enables repair/rehydrate workflows.
Benefit: SQL-friendly analytics and structured querying.
Why: complements semantic search with tabular and operational insight paths.
Benefit: long-horizon archival context for agent reasoning.
Why: deep memory context improves difficult recall beyond nearest-neighbor hits.
Benefit: resilient async delivery with retries, coalescing, and admission control.
Why: prevents sink instability from breaking ingestion reliability.
Benefit: bounded storage growth and observable runtime behavior.
Why: operational stability is required for learning retrieval to stay trustworthy.
The orchestrator's learning schema can only improve ranking if retrieval sources stay healthy, durable, and synchronized. This architecture makes that possible: Qdrant provides fast candidates, Letta supplies deeper RAG context, Mongo guarantees recovery, MindsDB adds structured recall, and guardrails keep the full loop from collapsing under pressure.
Best for local development and constrained laptops where stable memory services matter more than deep analytics.
Best for high-write workloads and richer retrieval where learning loops use every sink, including RAG through Letta.