One agent sets up the rest

Pick your first agent.

Add Engrams to one CLI agent. Then ask it to install and configure the others. Three commands, five platforms.

Claude Code

Anthropic's CLI agent. If you're already using it, start here.

Gemini CLI

Google's terminal agent. npm install, one config file, done.

Codex CLI

OpenAI's coding agent. Native MCP support via config.toml.

Run once in any terminal

Run the command above. Registers Engrams globally.

Start Claude Code. Say "boot" — your memory loads.

Say "Set up Gemini CLI and Codex with Engrams" — it does the rest.

Step 1 — Install

npm install -g @google/gemini-cli

Step 2 — Add to ~/.gemini/settings.json

Install Gemini CLI and create the config above.

Run gemini. Say "What are my Engrams projects?"

Say "Set up Claude Code and Codex with Engrams" — it does the rest.

Requires a Gemini API key from Google AI Studio (free). Set it as GEMINI_API_KEY environment variable.

Add to ~/.codex/config.toml

Add the config above to your Codex config file.

Start Codex. Say "What are my Engrams projects?"

Say "Set up Claude Code and Gemini CLI with Engrams" — it does the rest.

Then what?

Your AI stops starting over. Decisions you made last week, context from yesterday's session, project state across tools — it's all there when you start a new conversation. Switch between Claude, ChatGPT, Gemini, and Codex without losing a thing.

What one agent sets up for the others:
Claude Code → claude mcp add (one command)
Gemini CLI → writes ~/.gemini/settings.json
Codex CLI → writes ~/.codex/config.toml
Cursor → writes ~/.cursor/mcp.json

Try it — 60 seconds

Paste this into your agent after setup. It saves a memory, reads it back, and proves the loop works.

Paste into your agent

Save this to Engrams: "I'm testing Engrams for the first time." Then recall it back and show me what you found.

If it works, start a new session and ask: "What do you know about me?" — the memory carries over. That's the point.

Verify connection

Save a test memory and read it back in one click.

Web apps

Claude.ai, ChatGPT, and Lovable need manual setup since they don't have terminal access. Click to expand.

In Claude.ai, click your name → Settings → Connectors

Add custom connector → name it Engrams → paste this URL:

Click Connect. Done.

In ChatGPT, click + (bottom left) → Add MCP connector

Paste this URL:

Click Add. For best results, create a ChatGPT Project with Engrams attached and add this instruction: At the start of every conversation, call Engrams boot.

In Lovable, go to Settings → Integrations → Your MCP Servers

Click Add, paste this URL:

Done. Lovable reads and writes to the same memory as your other tools.

After setup

Memory loads automatically when you start a session and saves as you work. Private to you — no other user can access it.

Run verification tests →