What louvain is
Permissioned memory for company AI — every answer scoped to the person asking.
Your company already knows the answer. It is in a Slack thread from March, a document nobody linked, and the head of someone who has since changed teams.
louvain reads those sources and turns them into memory an AI can query — and it refuses to tell anyone something they were not already allowed to know.
The part that is different
Most tools that add memory to AI build one index and hope the prompt keeps the secrets. louvain resolves permissions before ranking: a question is answered against the subset of the corpus the person asking can already see. Two people can ask the same question, in the same deployment, on the same day, and correctly get different answers.
A correct refusal and a missing record look identical from outside. That is deliberate. Anything else leaks the existence of what it is hiding.
Start here
Quickstart — hosted
Sign in, connect a source, ask your first question.
Quickstart — self-hosted
Run the whole system yourself, including the model.
Understand the model
Worth twenty minutes before you connect anything real — the permission and provenance rules are what make the answers trustworthy, and they have consequences you should expect rather than discover.
Memory
Three tiers: typed claims, open-vocabulary memories, verbatim passages.
Permissions
How the fence is computed, and why it runs before ranking.
Provenance and time
Supersession, hedged claims, and what 'current' means per viewer.
Connect your sources
Connections
What a connection is, and how it becomes a permission boundary.
OAuth sources
Twenty-one platforms, and what each can tell louvain about who may see a thing.
Slack
Channels, identity linking, and corporate exports.
Push your own data
Send events from any system to the ingest endpoint.
Use it
Asking questions
What an answer contains, and why it sometimes refuses.
Administration
Members, roles, invites and SSO.
Limits and usage
What is metered and what happens at the ceiling.