Engram

An agent-agnostic memory layer. Capture facts about you and your work from any coding agent, review them on your terms, and recall them everywhere.

Engram runs locally, stores memories as plain Markdown you own, and speaks the Model Context Protocol so it works with Claude Code, Codex, opencode, and any MCP-capable client - driving cloud or local models (LM Studio, Ollama) alike.

Status: early development. The core engine and MCP server are being built in the open. APIs will change.

Why

Coding agents forget everything between sessions. Workarounds exist, but each is locked to one tool: every harness has its own memory, and none of them share. Engram is the shared brain - one local store that every agent reads from and writes to, with you as the gatekeeper.

Related MCP server: Agent Memory Bridge

How it works

   any agent ──remember()──▶ ┌───────────┐ ──auto-log──▶ memory-log.md
   (mid-task)                │  engram   │
                             │  capture  │ ──gate──▶ review queue ──you approve──▶ memory.md
   session transcripts ─────▶│  + bridge │
   (local-model harvest)     └─────┬─────┘
                                   │
   recall ◀── MCP resource ────────┤
   recall ◀── generated AGENTS.md ─┘

• Capture - agents call a remember tool mid-task, or Engram harvests durable facts from session transcripts using a local model.

• Review - low-risk facts are logged automatically; anything sensitive waits in a queue you approve. Nothing rewrites your curated notes without consent.

• Recall - every agent loads your memories through an MCP resource or a generated AGENTS.md / CLAUDE.md context block.

Supported clients

Quickstart

uv tool install engram          # or: pipx install engram

engram remember "I prefer pnpm over npm"
engram recall                   # list what engram knows
engram serve                    # start the MCP server for your agents

Wire it into an agent (Codex shown):

# ~/.codex/config.toml
[mcp_servers.engram]
command = "engram-mcp"

Design principles

• Local-first. Your memories never leave your machine. No telemetry.

• You own the data. Plain Markdown + YAML, git-diffable, no database lock-in.

• Human in the loop. Tiered writes: auto-log the trivial, gate the sensitive.

• Bring your own model. Any OpenAI-compatible endpoint extracts memories - cloud or local.

How it compares

Most memory tools are vector stores the agent writes to directly. Engram takes a different stance:

It federates capture across your agents, gates sensitive writes behind your approval, and keeps everything in a plain-text store on your machine.

Documentation

• Quickstart

• Architecture

• Security and privacy

• Adapters

License

MIT