Deployment¶

Engram can be deployed as a local daemon, a Docker container, or a production service with PostgreSQL and Redis.

Deployment Options¶

Option	Use Case	Complexity
Local daemon	Single developer, local AI agents	Minimal
Docker (single)	Isolated environment, easy setup	Low
Docker Compose	Production with PostgreSQL + Redis	Medium
Kubernetes	Multi-tenant, high availability	High

Local Daemon¶

The simplest deployment — runs engram as a background process on your machine:

pip install engram-mem
engram init
export GEMINI_API_KEY="your-key"
engram start

The daemon starts: - HTTP API server at http://127.0.0.1:8765 - Session watcher (if capture is enabled) - Background scheduler (cleanup, consolidation, decay)

engram stop     # Stop daemon
engram health   # Check status

Docker¶

See Docker deployment for full instructions.

Quick start:

docker build -t engram:latest .
docker run -e GEMINI_API_KEY="your-key" -p 8765:8765 engram:latest

Health Checks¶

All deployment targets should monitor these endpoints:

Endpoint	Purpose
`GET /health`	Liveness — returns 200 if process is alive
`GET /health/ready`	Readiness — checks episodic, semantic, and LLM subsystems

Kubernetes probe example:

livenessProbe:
  httpGet:
    path: /health
    port: 8765
  initialDelaySeconds: 10
  periodSeconds: 30

readinessProbe:
  httpGet:
    path: /health/ready
    port: 8765
  initialDelaySeconds: 15
  periodSeconds: 10

Resource Tier Monitoring¶

engram resource-status

Tier	Meaning
`FULL`	All features available
`STANDARD`	No LLM (API unavailable)
`BASIC`	Keyword search only
`READONLY`	Read-only mode

The system auto-recovers to FULL within 60 seconds of the LLM API becoming available again.

Backup and Restore¶

# Export snapshot
engram backup
# Creates: ~/.engram/backup-YYYYMMDD-HHMMSS.json

# Restore
engram restore ~/.engram/backup-20260306-013000.json

Snapshots include all episodic memories and semantic graph data. Schedule regular backups in production via cron or the Docker Compose setup.

Observability¶

Enable OpenTelemetry and audit logging for production:

audit:
  enabled: true
  path: /var/log/engram/audit.jsonl

telemetry:
  enabled: true
  endpoint: http://otel-collector:4317

Or via environment variables:

export ENGRAM_AUDIT_ENABLED=true
export ENGRAM_TELEMETRY_ENABLED=true