Your memory. Your database. Your environment.
Build your own AI memory on Memgraph and shape it to how you work. Persistent memory for agents, installed in one line, open to Cypher, yours to change.
curl -sSL https://install.memgraph.com/ai-memory | bash
- Claude Code
- Codex
- GitHub Copilot
- Cursor
- OpenCode
- Antigravity
- Grok Build
Today, your agent starts every session from zero.
Memory gives it continuity. The question is how much control you need over what happens next.
Decided for you.
The provider chose the storage, the schema and the extraction pipeline for its own reasons. You adapt to them, not the other way round.
Closed to you.
You cannot open the memory with a query, change how it is modelled or audit what an agent read.
Similar, not relevant.
Recall by similarity returns text that resembles the prompt. It cannot follow a fact to the session that stated it, or to the person who owns the system it concerns.
Cost you can’t shape.
Extraction and reconciliation run on the provider’s models, on the provider’s schedule, and you pay for every call.
One engineer only.
A team memory needs access that follows people. A hosted API does not know who your people are.
Build AI memory on Memgraph and shape it to your environment. See how it works under the hood.
All of it runs inside your walls.
Four steps, one database. Capture what the agent does, keep what matters, understand it after the session, serve it back next time.
Agent
Claude Code, Codex, Cursor or an Agent SDK app
Capture
Hooks · secrets redacted
memgraph-context-graph
hooks, MCP server, SDK, setup
Memgraph
Reconciliation
Runs locally, after a session
sessions
CI is red on test_login again. Any idea why?
It’s the shared token cache in auth-service again. Last time the fix was the fresh_token_cache fixture. Ana owns auth-service, if the TTL needs to change.
Yours. Fast. Token-light.
Yours.
Under your control.
Database, schema and extraction model run where you decide. Open it with Cypher. Audit what every agent read.
Fast.
Low latency.
Recall is one query against an in-memory graph, so memories come back in seconds per answer, measured below.
Token-light.
Fewer tokens spent reaching the answer.
A small local model extracts entities. The agent reads three memories, not 214 events.
Tokens are money you spend every day once agents are running. Tokens per answer is the one number here a finance team can read as is.
Answer score, tokens per answer and latency, measured on LongMemEval and BEAM.
See the resultsStart with one user. Grow into a team.
AI memory runs on Memgraph Community. One developer, multiple agents, one memory graph, nothing to buy.
Just you
- Stops starting from zero
- Runs on your laptop
- Open it with Cypher
- Nothing to buy
Your team
- One memory, many users
- Access follows people: role-based access control, with label- and property-level control underneath
- Your infrastructure, your costs
- Audit what every agent read
Sharing one memory across a team means not everyone should see everything. Enterprise makes each agent see only what its person is allowed to.
Same question. Different answer.
Search runs as the person the agent works for. What they can’t read never reaches the model.
- platform-eng
- everyone
- finance
Ana sees all 11 memories.Marco sees 8 memories. Nothing says three are missing.
Don’t trust our numbers. Run them.
LongMemEval-S
500 questions
- Answer score
- 87.4%
- Tokens per answer
- 5,863
- Latency
- 1.56 s
BEAM 100K
400 questions
- Answer score
- 63.1%
- Tokens per answer
- 5,865
- Latency
- 0.50 s
uv run context-graph-eval run <workload>
Give your agent a memory.
Install in one line.
curl -sSL https://install.memgraph.com/ai-memory | bash
- Claude Code
- Codex
- GitHub Copilot
- Cursor
- OpenCode
- Antigravity
- Grok Build
Already on Memgraph? Prefix with MEMGRAPH_HOST=your-host.
Share one memory, safely.
Start an Enterprise trial, join a JumpStart programme and we’ll build it with you, or bring your questions to Office Hours.
See how to set up AI memory on Memgraph.
Read the docsJoin the launch community call on October 29th. We install it, open it with Cypher and run the numbers in front of you.
