Engram

Knowledge corpus service with vector search - your AI's long-term memory.

Overview

Engram is a FastAPI-based service providing semantic search over a knowledge corpus using PostgreSQL with pgvector. It runs on GPU-enabled infrastructure for fast embedding generation with sentence-transformers.

Requirements

• Python: 3.11+ (3.12+ recommended for production)
• GPU: NVIDIA GPU with CUDA support
• Base Image: NVIDIA PyTorch 25.12 (for GPU deployment)
• Database: PostgreSQL 16 with pgvector extension

Deployment Architecture

Internet/Tailscale → Caddy (TLS termination) → Engram (port 8800) → PostgreSQL + pgvector

Components

• Caddy: Reverse proxy with automatic TLS via Step CA ACME

  • Terminates TLS at engram.your-domain.example.com
  • Proxies to Engram on localhost:8800

• Engram App: FastAPI service

  • GPU-accelerated embedding generation
  • Runs in NVIDIA PyTorch container
  • Uses BAAI/bge-large-en-v1.5 model (1024-dim embeddings)

• PostgreSQL: Vector database backend

  • pgvector extension for similarity search
  • Separate container on internal network

Certificate Management

TLS certificates are automatically managed via:

• Step CA: Internal ACME server at your-ca.example.com:9000
• Auto-renewal: Caddy handles certificate lifecycle
• Zero-trust: Mesh-only access via Tailscale

Quick Start

See deploy/README.md for deployment instructions.

Development

# Install dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Run locally
python -m engram.cli

License

MIT