Getting started

Engram runs in one of two deployment profiles, selected by the DEPLOYMENT_PROFILE environment variable. Both run on PostgreSQL alone (pgvector included) — pick the one that matches whether you need multi-tenant auth, then follow the matching tutorial.

Profile	DEPLOYMENT_PROFILE	External dependencies	Use it for
Lite	lite	Postgres only	A personal, single-user local memory server
Standard (default)	standard	Postgres only	Shared or production deployments — auth, API keys, tenancy

The legacy value enterprise is an accepted alias for standard. The old memory profile was removed — every profile now runs on Postgres.

Which page do I want?

• Quick start — the fastest path: one command sets up the lite profile and starts a running MCP server. Start here if you just want to try Engram.
• Installation — manual setup for both the lite and standard profiles, plus the command and infrastructure reference.
• Store your first memory — call create_memory, recall, and get_memory end-to-end against a running server.
• MCP client setup — connect Claude Desktop or Claude Code to your server.

Prerequisites (both profiles)

• Node.js 22.13.0 or newer with npm
• Git
• Docker and Docker Compose v2 — to run the bundled PostgreSQL container (image pgvector/pgvector:pg17). Or point DATABASE_URL at your own Postgres with the pgvector extension.
• Optional: pnpm on your PATH (the repository pins pnpm@11.5.0)

When pnpm is not installed, replace the leading pnpm in any command with npm exec --yes pnpm@11.5.0 --. For example:

npm exec --yes pnpm@11.5.0 -- install

Where to go next

Once your server is running, the how-to guides cover production deployment, backups, and observability, and the reference section documents every MCP tool and environment variable. To make Engram the shared primary memory for a fleet of AI coding agents, start at the agent memory contract.