Engram

What Engram does

Engram stores agent memory as a plain, git-backed folder of markdown files that you own. There is no database: the .md files are the source of truth, git is the durable store, and an in-memory index powers full-text search and a wikilink knowledge graph. You can point it at a fresh repository or an existing Obsidian vault, and agents reach it over one MCP endpoint.

Its focus is a team running several agents against one memory. Per-agent read and write tokens and a git audit trail record who wrote what. A fact can be marked superseded or expired, and search then withholds it and tells the agent what it skipped and why, so an agent does not quote a retired price or an old API doc. A dashboard lets you search the vault, see every agent's and teammate's changes with per-file diffs, jump back into recent notes and curate content. The README includes Railway and Render deploy buttons and a container image on GitHub.

Key features

• MCP server that agents use to read and write long-term memory
• Git-backed markdown vault with no database
• Superseded and expired facts are withheld from search, with a note on what was skipped
• Per-agent read and write tokens and a git audit trail
• Dashboard with search, per-file diffs and curation
• Obsidian-compatible vault with a wikilink knowledge graph

When to use it

• Giving several agents on a team one shared memory
• Stopping agents from quoting outdated prices or API docs
• Exposing an existing Obsidian vault to agents over MCP

Who it is for: Teams and solo users who run several AI agents and want shared, auditable markdown memory they host themselves.

How it fits with Hermes Agent

Hermes is named among the supported agents next to Claude Code and Cursor, and it connects through the standard MCP endpoint.

FAQ