Engram is a graph-oriented DBMS designed to store context for AI agents. It allows agents to share context across different large language models and frameworks. The system supports cross-platform synchronization, delta operations for context updates, and reinforcement learning mechanisms to improve context quality over time.
- Website
- https://engram.so[01]
- Source Code
- https://github.com/softmaxdata/engram[02]
- Developer
- Country of Origin
- CA
- Start Year
- 2026
- Project Types
- Commercial, Open Source
- Written in
- Python
- Embeds / Uses
- PostgreSQL, SQLite
- License
- MIT License
Engram is a graph-oriented DBMS designed to store context for AI agents. It allows agents to share context across different large language models and frameworks. The system supports cross-platform synchronization, delta operations for context updates, and reinforcement learning mechanisms to improve context quality over time.
History
Engram seeks to address limitations in AI agent memory management, such as context decay and isolation between models. The system draws inspiration from cognitive science concepts like associative recall and consolidation, as well as recent research on agentic context engineering and dynamic cheatsheets.
It is distributed via GitHub and offers a hosted service alongside self-hosted capabilities for cloud platforms.
Citations
4 sources- Engram — Persistent Memory for AI Agents engram.so
- GitHub - softmaxdata/engram: A brain-inspired, portable context database for AI agents · GitHub github.com
- GitHub - softmaxdata/engram: A brain-inspired, portable context database for AI agents · GitHub github.com
- https://github.com/softmaxdata/engram/commit/405459c8d1136e64478c909c0689883d4b859b19 github.com