Repository navigation
Releases
v0.6.0 - Stable Version
Chat UI works, teardown finishes
Fixed
- Open WebUI: Fixed model cache volume mounting (
/app/backend/data/store), disabled direct Ollama API calls, and enforced strict offline mode (OFFLINE_MODE=true). - Teardown: Resolved
make downhanging on ArgoCD finalizers, added time-bounded destroys, and implemented a final sweep for dangling containers/networks.
Docs
- Updated README to feature the working chat UI.
- Renamed asset filenames to match kebab-case conventions.
v0.5.5
v0.5.5
Lightweight CPU Inference & Deployment Resilience
All working stable platform
![]()
⚡
- Qwen2.5:0.5b Integration: Replaced the default CPU profile model (
llama3.2:1b) withqwen2.5:0.5b, reducing local memory footprint and accelerating response times on non-GPU setups. - Synced Orchestration: Updated LiteLLM routing rules and department default models to seamlessly target the new
qwen2.5backend.
🛡️
- Non-blocking Model Pulls: Configured timeouts for model preload init-containers, ensuring cluster services (ArgoCD, LiteLLM, Traefik) start cleanly even if internet bandwidth is limited.
- Bootstrap Job Backoff: Increased retry threshold for initialization jobs.
📁
- Open WebUI Persistence: Included PVC template for persistent user data across restarts.
- Script Executables: Updated execution bits for all environment setup and teardown scripts.
- Architecture Media Overhaul: Refreshed system diagrams and updated main repository README with new project banner.
v0.5.2
v0.5.1
v0.4.0
Complete security hardening and operational automation for the platform.
NetworkPolicies (progressive hardening):
- Enable NetworkPolicy in Postgres (datastore)
- Enable NetworkPolicy in LiteLLM (gateway)
- Enable NetworkPolicy in Ollama (inference) with new template
- All components validated with helm lint and template
Reloader for automatic secret rotation:
- Add Reloader operator in wave -15
- Configure ServiceMonitor for Prometheus integration
- Add secret.reloader.stakater.com/reload annotations:
- LiteLLM deployment (litellm, postgres, redis secrets)
- Postgres StatefulSet (postgres secret)
- Redis StatefulSet (redis secret)
- Completes External Secrets Operator flow from ADR-0007
Presidio enabled in GPU profile:
- Enable presidio in values-gpu.yaml
- Add ~2 GiB RAM and latency impact warning
- Monitor performance impact after deployment
Documentation:
- Update GOTO.md marking NetworkPolicies and Presidio as completed
- Add v0.4.0 release notes with testing and next steps
- Document potential connectivity and performance adjustments
Testing:
- Helm lint and template validation passes for all changes
- Platform validation passes for both GPU and CPU profiles
- NetworkPolicy templates verified
Part of v0.4.0 release focused on security and operations
v0.3.1
v0.3.0
v0.1.0
Primera versión que funciona de principio a fin: make up levanta el
gateway, la inferencia, el escalado y la observabilidad, todo por GitOps y a
coste 0.
Incluye: gateway con presupuestos por equipo (LiteLLM), inferencia local
con Ollama, autoescalado por RPS con KEDA, observabilidad GenAI (TTFT,
tokens/s, VRAM), guardrails de PII, cloud mockeada con LocalStack, y
Checkov/Trivy/Gitleaks en el pipeline.
Falta: UI de chat (usa el endpoint OpenAI-compatible), y con una sola
GPU no escalan réplicas : detalles en el README. #5
:)
