Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/xantorres/engram/llms.txt

Use this file to discover all available pages before exploring further.

Engram is a shared memory layer for your coding agents. Every session, agents like Claude Code, Codex, and opencode forget everything they learned. Engram fixes that — one local store every agent reads from and writes to, with you as the gatekeeper for anything sensitive.

Quickstart

Get Engram installed and your first memory captured in under five minutes.

Installation

Install from source with uv, pipx, or directly from a local clone.

Core Concepts

Understand the memory model, risk tiers, and the capture-review-recall lifecycle.

Wiring Agents

Connect Claude Code, Codex, opencode, or any MCP-capable client to Engram.

How Engram works

Engram has three phases that keep your knowledge base useful and trustworthy:
1

Capture

Your agent calls the remember MCP tool mid-task, or Engram harvests durable facts from past session transcripts using a local model (LM Studio, Ollama) or any OpenAI-compatible endpoint.
2

Review

Low-risk fact kinds (preferences, tooling choices, project details) are logged automatically. Sensitive kinds — fiscal, identity, health, people — wait in a queue you approve. Nothing rewrites your memory without consent.
3

Recall

Every agent loads your memories through an MCP resource (memory://recall) or a generated AGENTS.md / CLAUDE.md context block. All agents share the same store.

Key capabilities

Local-first storage

Plain Markdown + YAML on your machine. No cloud, no telemetry. Back it up with Git, Dropbox, or any file sync you already use.

Tiered write safety

Tier 1 facts auto-log. Tier 3 (sensitive) facts require explicit --confirm. Conflicts always queue for review.

MCP-native

Ships an MCP stdio server (engram-mcp) with remember and recall tools and a memory://recall resource any client can load.

Transcript harvesting

Mine session transcripts from Claude Code, Codex, or opencode with any OpenAI-compatible extractor model.

Full audit trail

Atomic writes, append-only audit.jsonl, single-step undo tokens, and a doctor command for staleness and conflict reports.

Any agent, one store

Works with Claude Code, Codex, opencode, and any MCP-capable client — cloud or local models alike.

Supported agents

AgentCaptureRecall
Claude CodeMCP tool + transcript harvestMCP resource + CLAUDE.md block
CodexMCP tool + transcript harvestMCP resource + AGENTS.md block
opencodeMCP tool + transcript harvestMCP resource + AGENTS.md block
Any MCP clientMCP toolMCP resource
Engram is in early development. The core engine and MCP server are being built in the open. APIs may change before the 1.0 release.

Build docs developers (and LLMs) love