what is engram fr?

Engram is a self hosted memory system for AI coding assistants and agents, such as Claude Code, Cursor, and others that support the Model Context Protocol, a standard way for AI tools to connect to external data sources. It gives those agents a shared, long term memory they can both read from and write to, built on top of a plain folder of markdown text files that the user owns and stores in git, rather than a database or a specialized vector search system. The problem it addresses is that AI agents normally forget everything between sessions, and when they do recall old information, such as an outdated price or a document that changed months ago, they tend to state it with full confidence even though it is no longer true. Engram treats whether a piece of information is still accurate as an explicit, written property: a fact can be marked as superseded or expired, and when an agent searches, outdated entries are withheld from the results, with a note explaining what was skipped and why. Because the memory lives in git backed markdown files, every change any agent or teammate makes is tracked with a full history, including per file differences, and each agent can be given its own separate read and write access token. A companion dashboard lets a person search the shared memory directly, watch recent changes as they come in, browse a knowledge graph built from links between notes, and jump back into any note to read or edit it, all while agents continue reading and writing to the same underlying files through one shared connection point. The project is explicit that it has opinions about how memory is stored, meaning git backed markdown with no database, but no opinions about what content goes into it: it can point at a brand new folder or an existing Obsidian notes vault with no import step needed. It is released under the MIT License and can be deployed with Docker or one click templates to hosting services such as Railway or Render.