Free & Open Source · npm package

Give your AI a memory.

One script. A handful of markdown files. Your AI stops forgetting what you talked about yesterday. Works with Claude, ChatGPT, Gemini, Cursor, and Antigravity.

Before and after.

The difference between raw AI and AI with memory.

Session 12 — Without Engram

Last time we discussed the caching strategy—
I don't have context from previous conversations. Could you summarize what you've discussed so far?

Session 12 — With Engram

Reading STATE.md… Session 12. Last session we finalized the Redis caching approach (Decision 7 in DECISIONS.md) and opened a thread on rate limiting — two open questions: token bucket vs. sliding window, and what the default rate should be. Pick up there?

Context engineering, not brute force.

The AI reads two small files at session start — not 50 pages of logs. A layered retrieval model that scales to hundreds of sessions without hitting context limits.

HotSTATE.md + ENGRAM.mdEvery session~1,000–3,000
WarmHANDOFF.mdSession start~500–1,500
ColdENGRAM-INDEX.mdOn-demand grep~200–800
ArchiveENGRAM-LOG.mdReconcile only15K–200K

Session cold-start: ~1–3K tokens. At 50 sessions, a raw log would be 80,000–200,000.

No database. No API. No plugins. Just markdown and a protocol.

What Engram does.

Layered Memory

Boots in ~1–3K tokens. Scales to 200+ sessions without hitting context limits. The AI reads a summary, not the full history.

MCP Server + CLI

11 MCP tools, 4 resources, 9 CLI commands. Auto-configures for Claude Desktop, Cursor, and Antigravity with a single flag.

5 Project Templates

default, research, software, writing, startup — domain-specific workstreams and file structure out of the box.

Live Dashboard

Session timeline, decision log, workstream tracker, agent activity. Run engram-watch.sh --daemon and the dashboard auto-refreshes.

MCP Server + CLI

The npm package includes a Model Context Protocol server that gives AI tools direct read/write access to your engram files — plus a full CLI for manual operations.

MCP Tools (11)

engram_status · engram_log_exchange · engram_checkpoint · engram_search · engram_decisions · engram_reconcile · engram_handoff · engram_rotate · engram_update_state · engram_update_summary · engram_add_decision

MCP Resources (4)

engram://state · engram://summary · engram://decisions · engram://agents

CLI Commands (9)

engram setup · engram serve · engram status · engram checkpoint · engram search · engram decisions · engram reconcile · engram handoff · engram rotate

Auto-configures for your AI client

npx engram-protocol setup --all        # Claude Desktop + Cursor + Antigravity
npx engram-protocol setup --claude-desktop
npx engram-protocol setup --antigravity

See your project's memory.

Run ./update-visualizer.sh once — or ./engram-watch.sh --daemon to keep it live. Open VISUALIZER.html in any browser. No server. No internet required.

Engram dashboard overview showing sessions, decisions, workstreams, and AI agents
Engram session timeline showing chronological sessions with tags and AI model attribution

One command. Three minutes.

Install. Run. Your AI has a memory.

# Setup with npx (recommended)
npx engram-protocol setup --name "My Project"

With options:

npx engram-protocol setup --name "My Research" --author "Jane" --template research --all

Alternative: Shell wrapper (interactive prompts, backup/restore, --uninstall)

curl -O https://raw.githubusercontent.com/ecomxco/engram/main/init-engram.sh && chmod +x init-engram.sh && ./init-engram.sh

Open the folder in Claude Code, Cursor, or Antigravity — engram activates automatically via .mcp.json. No config editing.

One command. Your AI never forgets again.

Free forever. MIT licensed. No email. No account. npm install or curl — your choice.

Built for operators.

npm package · MCP server · 9 CLI commands · 5 templates · Any AI platform · MIT License