Graphiti Memory
Graphiti Memory gives Kaji the ability to remember information across sessions. Instead of starting fresh every conversation, Kaji can store facts, preferences, and context that persist over time — making it more useful the more you use it.
What Kaji can do
- Store notes, facts, and preferences for later recall
- Remember details from previous conversations
- Build up a knowledge graph of your team's context
- Recall stored information relevant to a current task
Example prompts
Remember that our main customer contact at Acme Corp is alice@acme.com and she prefers executive summaries
What do you remember about our Q1 roadmap decisions?
Store this for later: our data warehouse is named "prod_dwh" in Dremio
Do you recall anything about the authentication migration project?
Forget everything you know about the old pricing model
How memory works
Graphiti uses a knowledge graph to store memories — entities (people, projects, concepts) and the relationships between them. This allows Kaji to make connections and surface relevant context intelligently.
For example, if you tell Kaji "Alice is the lead on Project Apollo" and later ask "who is working on Apollo?", Kaji can recall this from memory.
Tips for business users
- Be explicit when you want Kaji to remember something: "Remember that..."
- Memories are scoped to your organization — they are shared within your Kaji environment, not across different organizations
- You can ask Kaji to "forget" specific information if it's no longer accurate
- Memory works best for persistent facts (team members, project names, preferences) rather than transient data
Configuration
Graphiti Memory requires a Neo4j database for storing the knowledge graph.
| Variable | Description |
|---|---|
NEO4J_URI | Neo4j connection URI for the memory graph |
NEO4J_USERNAME | Database username |
NEO4J_PASSWORD | Database password |
In standard Kaji environments on Shakudo, this is pre-configured automatically.