feat: self-hosted LLM backend — point all workflows at the Jetson (Ollama, model=coder)

The Jetson Orin Nano (JetPack 7.2, Ollama on GPU) is live, so ci-ai runs fully
self-hosted: no Anthropic key, no cloud, no per-PR cost. AI_PROVIDER=ollama,
AI_BASE_URL=http://192.168.178.103:11434/v1. The Pi runner reaches it LAN-direct
(SNAT to .48, allowed by the Jetson UFW). Removed the dead ANTHROPIC_API_KEY
secret declarations.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
2026-06-12 21:58:07 +02:00
parent eb0437b0ed
commit 2ae14a51e4
7 changed files with 26 additions and 40 deletions

View File

@@ -3,7 +3,7 @@
# Läuft auf dem Pi-Runner (capacity 1, ~3-4 GB RAM) -> Memory-Cap + schlankes
# python-Image. KEINE JS-Actions (kein node nötig): ci-ai wird manuell geklont,
# git.eulernest.eu löst per Runner-add-host auf die LAN-IP auf. Ohne
# ANTHROPIC_API_KEY überspringt das Skript den KI-Schritt sauber (Job bleibt grün).
# Das KI-Backend ist die Jetson-Ollama (self-hosted, LAN-direkt, kein Key).
name: AI PR Review
on:
@@ -26,15 +26,10 @@ jobs:
- name: KI-Review (überspringt sauber ohne Key)
env:
AI_PROVIDER: anthropic
AI_MODEL: claude-3-5-haiku-latest
ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
AI_PROVIDER: ollama
AI_MODEL: coder
AI_BASE_URL: http://192.168.178.103:11434/v1
GITEA_TOKEN: ${{ github.token }}
GITEA_API: https://git.eulernest.eu/api/v1
run: python /opt/ci-ai/scripts/ai_review.py
# Jetson-Umstieg später (kein Code-Change, nur diese env):
# AI_PROVIDER: ollama
# AI_MODEL: qwen-light
# AI_BASE_URL: http://192.168.178.20:11434/v1
# JETSON_API_KEY: ${{ secrets.JETSON_API_KEY }}