Persistent memory for conversational AI
iLAB Memory is a persistent memory library for AI agents. It runs inside your application's process — like SQLite or an ORM — so your chatbot, voice assistant, or agent can remember a user across sessions, days, and restarts. No remote server, no vector database, no extra infrastructure.
Why iLAB Memory
Embeddable, zero ops
Imports as a Python library, persists locally to SQLite. No SaaS to deploy, no service to operate, no token leaving your stack.
SQLite + FTS5
Full-text search built-in via SQLite's FTS5 virtual tables. No vector database needed for v0.1 — keyword search is fast and predictable.
Privacy-first
Wrap secrets in <private>...</private> and they are stripped before hashing, deduping, or persistence. Sensitive content never reaches storage.
Topic_key upsert
Idempotent saves on the same topic_key — no duplicate explosion when the same topic evolves across turns.
Sessions with scoring
Recency and revision scoring built-in. Search and context use distinct formulas so each surface optimizes for its own intent.
Dual API surface
Use it as a Python library, expose it over HTTP REST (FastAPI), or plug it into IDEs and agents over MCP stdio. Both APIs serialize the same Pydantic models.
Who it's for
Bot developers
Building a Teams, Slack, or WhatsApp bot that needs to remember each user across days.
App developers
Embedding memory into a Tauri or Electron desktop app where the user owns their data locally.
Agent developers
Building autonomous Python agents that need persistent context between runs.
What it's NOT
- Not a vector database. v0.1 ships keyword search via FTS5 — semantic search via embeddings is on the roadmap.
- Not a remote SaaS. You embed it in your app; you own the database file. There is no hosted service to sign up for.
- Not an agent framework. It's the storage layer. You bring the LLM, the prompt orchestration, and the business logic.
Next steps
Quickstart
Go from zero to your first saved-and-retrieved observation in under 5 minutes.