Minimum (demo / laptop)
| Requirement | Notes |
|---|
| Docker Compose | Category stack |
| CPU / RAM | ~4+ CPU, 8+ GB RAM recommended for full stack |
| Disk | Clones + Postgres + graph; plan for workspace growth |
| LLM API access | OpenAI / Anthropic / SpaceXAI / compatible |
Production
| Requirement | Notes |
|---|
| Postgres | Required job + product SoT (DATABASE_URL) |
| Kubernetes (typical) | Helm chart: API, workers, UI |
| Keycloak (or OIDC IdP) | Recommended identity SoT |
| Graph backend | Neo4j or JanusGraph shared across workers |
| SCM credentials | GitHub App preferred for GitHub |
| Secrets management | API keys, DB URL, webhook secrets |
| Outbound HTTPS | To LLM providers and SCM APIs |
Optional
| Component | When |
|---|
| NATS / RabbitMQ / Pulsar | KEDA queue-depth autoscaling |
| Docker-in-Docker / Docker socket | Strict sandbox for agent tools |
| Object storage | Future WORM audit (not required for v1) |
Skills on your team
Someone comfortable with containers or Helm, OIDC/Keycloak basics, and reading worker logs will unblock install quickly. Deep LLM expertise is optional — start with one provider and the default specialist matrix.