Engram

Engram is JamJet's memory layer for AI agents — a temporal knowledge graph that stores conversational context, facts learned over time, and the relationships between them. It's used by JamJet workflows to give agents long-term memory across sessions.

Engram has its own dedicated docs at java-ai-memory.dev. This page is a stub pointing you there. Engram is also available as:

• Rust crate: jamjet-engram on crates.io (0.5.0)
• Python: jamjet-engram on PyPI (0.1.0)
• Java: engram-spring-boot-starter on Maven Central (0.2.0)
• Docker: ghcr.io/jamjet-labs/engram-server for the standalone server
• MCP: io.github.jamjet-labs/engram-server on the official MCP registry

Why Engram?

Most agent memory libraries store transcripts. Engram stores temporal events with relationships — you can query "what did the user tell me about their dog last month?" or "what tools has this agent used in the past 24 hours?" without scanning a flat history.

Benchmarks

Engram scores 88.8% on LongMemEval-S (proxy run, 500 questions) — competitive with proprietary memory products. See java-ai-memory.dev/benchmarks for details.

Get started

The full quickstart, API reference, and integration guides live at java-ai-memory.dev. Come back here when you want to combine Engram with the JamJet runtime or JamJet Cloud governance.