Secure, multi-tenant AI platform for Fireball Industries — LLM gateway, RAG pipelines, and agentic workflows on K3s.
| Component | Image | Purpose |
|---|---|---|
| LiteLLM | ghcr.io/berriai/litellm |
OpenAI-compatible API gateway with model routing |
| n8n | docker.n8n.io/n8nio/n8n |
Workflow orchestrator for RAG and automation |
| pgvector | ankane/pgvector |
PostgreSQL + vector embeddings for RAG |
| Valkey | valkey/valkey |
Redis-compatible cache and queue backend |
| OpenZiti | openziti/ziti-edge-tunnel |
Zero-trust overlay network (EmberNet Flux) |
| ZeroTier | zerotier/zerotier |
Admin interface access (sidecar) |
| Name | Model | Role |
|---|---|---|
nemotron-120b |
Nemotron 3 Super 120B | Main reasoning engine |
nemotron-4b |
Nemotron 3 Nano 4B | Routing, classification, fast tasks |
.\deploy.ps1
wsl -d Ubuntu k3s kubectl get pods -n ember-ai- Developer Onboarding — Setup, first API call, troubleshooting
- Developer Guide — Full architecture and component reference
- Architecture — Design principles
- Infrastructure — K3s prerequisites and setup
- Integration — Data flow between components
All secrets injected via GitHub Actions from GitHub Secrets at deploy time.
Proprietary — Fireball Industries LLC