Skip to content
Engram Docs

Quick start

The fastest path to a running Engram server is the lite profile, set up by the bundled installer. It is single-user with no auth to configure, and it persists memories to PostgreSQL just like standard.

You need Node.js 22.13.0+ and Docker (for the bundled PostgreSQL).

1. Clone, install, run

Section titled “1. Clone, install, run”
Terminal window
git clone https://github.com/osirison/engram.git
cd engram
./scripts/install.sh --start

The installer copies .env, installs dependencies, starts PostgreSQL, runs migrations, builds, and — with --start — launches the MCP server on http://localhost:3000. It is idempotent, so you can re-run it any time. Drop --start to have it print the start command instead of launching.

Prefer to wire each step yourself? See Installation.

2. Verify the server is up

Section titled “2. Verify the server is up”

In a second terminal:

Terminal window
curl http://localhost:3000/health

The health response reports ok with the process and database indicators.

3. Optional: local embeddings for hybrid recall

Section titled “3. Optional: local embeddings for hybrid recall”

The default embedding provider is Ollama — local, no API key. If an Ollama server is running (ollama.com/download) with the default model pulled, hybrid semantic recall works with zero extra configuration:

Terminal window
ollama pull nomic-embed-text

If Ollama is unreachable, the server degrades gracefully (memories are stored without vectors and can be backfilled by a later reindex). No Ollama at all? The deterministic hash provider gives stable — but not semantic — similarity without any model:

Terminal window
DEPLOYMENT_PROFILE=lite \
EMBEDDING_PROVIDER=local \
  pnpm --filter mcp-server dev

DEPLOYMENT_PROFILE is read at module-load time. If the server exits immediately, confirm the variable is set in the same shell that runs the build / dev commands — changing it after the process boots has no effect.

Next steps

Section titled “Next steps”