INFRASTRUCTURE

Networking, scaling limits, and dependencies for RUNE.

External Dependencies

• Vast.ai API: For provisioning GPU compute. Requires VAST_API_KEY.
• Ollama API: For model inference. Standard port 11434.
• S3-Compatible Object Store: For long-term result persistence (SeaweedFS, AWS S3).
• Kubernetes API: For HolmesGPT to query cluster state (Pods, Events, Logs).

Networking

• Port 8080: RUNE API server (internal or public).
• Port 11434: Ollama server.
• Port 8333: SeaweedFS S3 API.

Scaling Limits

• Concurrency: SQLite-backed job storage is simple and reliable for single-node or single-pod deployments, but it limits high-volume concurrent writes and does not provide a shared store for multiple rune-api replicas. The accepted PostgreSQL direction is documented in ADR 0006.
• Vast.ai Limits: Subject to provider availability and account limits.
• Model Size: Limited by target machine GPU memory (VRAM).

Security Boundaries

• HolmesGPT: Runs with read-only RBAC by default in target clusters.
• RUNE API: Uses tenant-scoped tokens for authorization.