for engineers

Give your AI agents a memory that lasts.

Lumen — persistent, governed, self-hosted memory for LLM agents.

the reset problem

Every session, your agent starts from zero.

Context evaporates

Decisions, conventions, and history — gone next session.

It guesses

Missing a fact, the model invents one. Confidently.

wire it in

Event hooks. No model changes.

SessionStart→ on prompt · ambient recall→ Stop · save + summarize

Works with any agent or LLM — drop it in, no fine-tuning.

one small API

Remember. Recall. Replay.

remember("prod is kube-proxy-free Cilium", scope) # save
recall("how is networking set up?") # lexical + vector + graph
timeline(since, until) # episodic context

Exposed as MCP tools + REST. Use it from Claude, your agents, or CI.

retrieval by fusion

Keyword, meaning, and relationships — ranked together.

Postgres · lexical Qdrant · vector Neo4j · graph

Queried in parallel, fused into one ranked recall — answers from your record, not vibes.

the stack

Built to run on your infrastructure.

Go Postgres Qdrant Neo4j MongoDB MinIO Redis MCP REST Kubernetes
governed + sovereign

Trustworthy by design — and entirely yours.

✓ Self-hosted — your infra, your data, no lock-in
✓ Verify-before-assert: source first, guess last
✓ Volatility flags on live facts
✓ Continuous correction from its own errors

Stop re-explaining your codebase to your AI.

Self-hosted. Governed. Yours.

Lumen — by StricklySoft  ·  stricklysoft.com/lumen