A pile of claims
Alice lives in Paris 2024
Alice lives in Berlin 2025
Alice lives in Lisbon 2026
Three observations. The current location still needs to be resolved.
Structured memory for AI agents
A changing world.
A memory that
keeps up.
Give your agents a shared record of people, projects, and decisions. When new information arrives, ConceptMem updates memory under your rules and keeps the history behind each change.
Connect throughMCPTypeScript SDKHTTP API
01 / OBSERVATION
“Alice moved from Berlin to Lisbon last month.”
ONTOLOGY + POLICY
person:alicePerson · current state
CURRENTlocationBerlinLisbon
roleStaff engineer
sourceAgent observation
✓
One current answer. Full provenance.Every previous value stays traceable.
01 / THE PROBLEM
More context is not
better memory.
A customer changes their preference. A project gets a new owner. A deadline moves. Your agent needs to know which information still applies.
Keeping every message preserves the conversation. Consolidating updates gives your agent a current record it can query, with previous values available when it needs the history.
One current state
CURRENT LOCATIONLisbonconfidence · 0.94
↳ Berlin and Paris remain in history
Query the current location. Inspect earlier values separately.
02 / THE SYSTEM
Memory with rules.
Define your model, govern updates, and retrieve the accepted record.
Model the world
Define classes, predicates, cardinality, and typed relations. Choose flexible validation or strict boundaries for each class.
↗Consolidate change
Send observations as proposals. Duplicates collapse, new values supersede old ones, and policy can hold changes for review.
↗Retrieve current truth
Look up entities, search facts, and traverse relations without replaying a lifetime of contradictory context.
↗03 / IN YOUR WORKFLOW
Remember what
the next task needs.
Start with a small part of your world that changes often and matters across conversations.
Customer context
Keep preferences and relationships up to date as new information arrives.
ExampleA preferred contact channel changes from phone to email.
Project continuity
Give planning and delivery agents a shared view of owners, milestones, and status.
ExampleA new deadline supersedes the previous date.
Team knowledge
Connect people, responsibilities, and projects through typed relationships.
ExampleA project gains a new owner; the change remains traceable.
04 / CONNECT
A memory connection.
Useful tools.
Connect ConceptMem as an MCP server. The same governed memory becomes available to every compatible agent and IDE.
Explore MCP →conceptmem.config.json•••
{
"mcpServers": {
"conceptmem": {
"url": "https://your-host/api/mcp",
"headers": {
"Authorization": "Bearer cmk_…"
}
}
}
}memory_insertmemory_searchmemory_getmemory_history
BEFORE YOU START
A few things
worth knowing.
What ConceptMem does, and how it fits your agent.
How is this different from storing conversation history?
Conversation history preserves what was said. ConceptMem consolidates observations into entity records, separates current facts from previous values, and records the provenance of changes.
Does it replace my document search?
ConceptMem focuses on structured, changing knowledge. It offers text search, exact lookups, and relationship traversal. You can use it alongside a document retrieval system.
Do I need to model everything first?
Start with a few classes and fields, or use the starter ontology. Flexible classes allow undeclared fields; strict classes refuse them. Add structure as your needs become clearer.
What happens when two observations disagree?
The effective class and field policy determines the outcome. Updates can supersede older values, use confidence-weighted arbitration, be rejected, or be held for review.
How do I connect my agent?
Create a workspace access key and follow the Connections guide for MCP. Applications can also connect through the TypeScript SDK or HTTP API.
Give your agents a memory
they can trust.
Define the world. Submit observations. Retrieve what is true now.
Create your workspace ↗