Give your AI a memory that lasts.
Context that persists. Built for production. Membrick helps an assistant remember useful preferences, decisions, and history across interactions, then brings the right context back when it matters.
A new name: Membrick
Previously known as Lint-AI, the project is now Membrick. Version 0.5.0 moves bundled provider integrations to a resident local IPC service. The standalone HTTP API remains available, and the legacy lint-ai command remains supported.
From one conversation to a lasting relationship¶
On its own, an AI model only has the context it can see right now. Membrick adds a memory that stays between conversations. As an agent saves new information or feedback, that memory can be updated and used in future interactions. The model does not need to be retrained; it gets better context to work with.
Remember the person and the work
Carry useful preferences, ongoing tasks, and past decisions from one interaction to the next. Your assistant can pick up where you left off instead of asking you to start over.
Learn from what changes
When new information or feedback updates a memory, future conversations can use it. Membrick keeps the earlier history too, so an old answer does not quietly become the current one.
Add memory without building it all yourself
Membrick handles saving, updating, and finding memories through one persistent service. Use the integrations and APIs to add memory to an agent without building a separate memory system from scratch.
What lasting memory makes possible¶
Connect Membrick to the experience you are building:
AI assistants
Remember preferences and past conversations to make each new interaction feel more personal and consistent.
Customer support
Bring relevant case history and past resolutions into the next support conversation, so people do not have to repeat themselves.
Autonomous systems
Carry goals, task progress, and previous outcomes across runs, so an agent can continue work with the state it has already built.
The same memory layer can support productivity tools and other applications that benefit from remembering user preferences and past activity. Connect these applications through the HTTP API, MCP, Python, or Rust interface.
Memory where your agents already work¶
Use the same project memory inside supported agent tools. Information captured in one can be available to the others, so people can keep using the tools they already know.
Built for dependable use¶
Membrick keeps memory in a persistent store and provides a shared service for agents and applications. It supports user and session scoping, durable writes, and retrieval that can account for newer corrections while preserving history. Run it locally or connect through its HTTP and MCP servers. A dashboard helps you inspect memory and agent activity. Bundled Rust integrations communicate with their resident memory owner over local IPC; see the transport architecture.
Benchmark highlights
These are the latest release-candidate results across all 500 LongMemEval-S questions, using the heuristic NER backend, structured-fact retrieval, a single index, and no embeddings. Fractional Recall reports how many relevant sessions were recovered; any-hit Recall reports whether at least one relevant session appeared in the results. See the full benchmark results for MRR, NDCG, search timing, methodology, and reports. HTTP throughput is measured separately and documented on the HTTP server benchmarks page.
Agent integration results and their limits are listed on the Performance page. The latest Claude comparison covers three repetitions of one scenario in each of three arms; it reports a 75% drop in fresh input tokens, a 79.0% to 79.6% drop in overall tokens, and a 16.0% to 18.5% drop in overall tokens per fresh input token. This is diagnostic evidence for that fixture and scenario, not a general performance claim.
Context that persists.