bg-mem gives language-model tools a durable, searchable long-term memory that you own and run yourself. Facts, decisions and context captured in one session are available to the next, to a different model, or to a different tool, without depending on any vendor's memory feature.
What it does¶
- Remember and recall: save atomic facts and search them semantically from any LLM client that speaks the Model Context Protocol.
- Verified facts: entries can be marked as authoritative so recall can prefer what has been checked.
- Works off-site: a portable mirror answers searches when the primary store is unreachable, and anything saved offline is queued and synced back later.
- Capture raw context: ingest unstructured conversation text as well as curated facts.
- Self-hosted: your data stays on your own infrastructure.
Availability
bg-mem is in use internally at Black Grid. Details will be published here as it matures.