01 · Remember
Facts, saved as they come up
People, decisions, dates, and preferences, one sentence each. The connector instructs the assistant to save them as they come up and to prefer this store over its built-in memory.
Memory & knowledge base for AI
Long-term memory for AI assistants
Store the facts, decisions, people, and documents an AI assistant should carry between chats. Ask what was current on any past date. Keep versioned documents, and have the assistant fetch the current version.
Works with Claude and other MCP clients
Connect with a URL and an API key
Your data is kept in its own graph with its own API keys
Current information
Keeping an assistant’s information current
Chat history is not kept between conversations, and copied files go out of date. InvariantDB stores facts and documents with their history so an assistant can fetch the current version.
How it works
Three operations
Save facts, look them up as of a date, and keep documents current. The assistant calls these operations itself.
02 · Recall, including past dates
“What was true on March 5?”
Every fact records when it became true and when it was saved. Ask for what is current, or for what was in effect on a past date.
03 · Versioned documents
One current version, with history
Save a document with a stable name; each revision supersedes the last, and earlier revisions stay readable. The connector instructs the assistant to fetch the current version before relying on a document.
Connect
Connect an AI client
- 1Create a workspace. Create it in the dashboard and copy its name and API key.
- 2Add the server. In your AI client, add a remote (HTTP) MCP server with the URL
https://invariantdb.com/graphs/<your-workspace>/mcp. - 3Approve. When the client opens the approval page in your browser, paste the API key and approve. If your client does not support OAuth sign-in, add the header
Authorization: Bearerfollowed by the key instead.
# MCP server URL (replace <your-workspace>) url = "https://invariantdb.com/graphs/ <your-workspace>/mcp" # then ask your assistant, for example: "remember that Rev 12 is the current pricing policy" "what did we decide on pricing in March?" "pull up the onboarding doc"
Who it's for
For people who use AI assistants and for people who build them
Knowledge workers & teams
Your AI, with a real memory of your world
- Frameworks, templates, and profiles that stay the current version everywhere.
- Decisions and context that carry from one conversation to the next.
- Organize it in your own words — your categories, not a generic schema.
- Nothing technical to run; the assistant handles the saving and recalling.
Developers & AI builders
A memory + knowledge-graph backend over MCP
- Drop-in memory for agents — save, recall, and point-in-time reads.
- Extract a queryable knowledge graph from documents.
- Full graph + search when you need it (Cypher, full-text, vector).
- A separate graph and API keys for each workspace.
Get started
Set up memory for your assistant
Request access, then create a workspace and connect your first client.