Context storage for AI agents.
Your agents store user context and recall it as a prompt-ready block. Every user gets their own isolated store, created on first write and built on LodeDB, the open-source embedded vector engine.
1from lodedb.cloud import Client23client = Client()4memory = client.store("user-42")5memory.add("likes hiking near Seattle")6context = memory.context_block("plan my weekend")One store per user, created on first write. Recall returns a prompt-ready block.
Name a store per user
A store name is the whole setup. The first write provisions it. No create step, no collections to manage.
Agents add memories
Writes are concurrent and read-your-writes. Attach TTLs to expire, or agent_id and run_id for provenance.
Recall in one call
Pass the raw message to get a prompt-ready context block, or point any MCP client at the store's endpoint.
Support agents
Voice and chat agents recall the relationship, not just the ticket.
“Still nothing back on that cancelled flight.”
“Your refund cleared this morning. Confirmation is in your texts.”
Shopping assistants
A correction made once holds across every future session.
“Too flashy. I never wear logo prints.”
“Twelve plain linen picks, all under €120.”
Coding agents
Memory written in one tool is recalled in another. Any MCP client, no app code.
“Stop using pip. This repo is uv-only.”
“Set up with uv sync, pip never touched.”
One database per user
Isolation is structural. Delete the store and the user is forgotten.
Prompt-ready recall
Pass the raw message, get back a fused context block in one call.
Hosted MCP per store
Every store is an MCP endpoint. Agents connect with one command.
A store is just LodeDB. Pull it down and open it locally.
The store IS the open-source on-disk format. Pull one to disk and open it with the embedded engine. The read API is identical. No export step, no proprietary conversion.
lodedb cloud pull orecloud://acme/production/user-7d2f ./user-7d2f <index-key>
# open locally, same read APIdb = lodedb.LodeDB(path="./user-7d2f")