PowerOn/service-llm-private

No description

Find a file

ValueOn AG b15d283941 All checks were successful Deploy LLM Service / deploy (push) Successful in 23s Details cp adapted to 2026 poweron		2026-06-09 09:54:11 +02:00
.forgejo/workflows	fix: use printf for SSH key to preserve trailing newline	2026-05-24 03:22:38 +02:00
deploy	ssl	2026-02-06 14:16:12 +01:00
docu	logging	2026-06-02 09:42:04 +02:00
static	initial	2026-02-06 00:30:52 +01:00
templates	Fix mixed content: route all requests through backend API	2026-02-06 14:35:38 +01:00
.gitignore	Initial commit	2026-02-06 00:24:28 +01:00
app.py	cp adapted to 2026 poweron	2026-06-09 09:54:11 +02:00
config.ini	Increase rate limits and add Retry-After header on /api/analyze	2026-03-29 23:29:23 +02:00
config.py	cp adapted to 2026 poweron	2026-06-09 09:54:11 +02:00
README.md	logging	2026-06-02 09:42:04 +02:00
requirements.txt	fixes req	2026-02-06 10:35:01 +01:00
routeApi.py	cp adapted to 2026 poweron	2026-06-09 09:54:11 +02:00
routeOpenAi.py	cp adapted to 2026 poweron	2026-06-09 09:54:11 +02:00
routeWeb.py	cp adapted to 2026 poweron	2026-06-09 09:54:11 +02:00
start-python.bat	integrated privateLLM	2026-02-06 10:27:06 +01:00
start-python.ps1	integrated privateLLM	2026-02-06 10:27:06 +01:00

README.md

Private LLM - für Neutralisierung

KI-Dokumentenanalyse und Text-Neutralisierung mit lokalen Ollama-Modellen.

Features

Text-Neutralisierung für den Gateway (kein Datenabfluss an externe APIs)
Rechnungen, Belege, Bankauszüge analysieren (Vision)
Handschrift erkennen
PDF-Support (PyMuPDF)
100% lokal auf Schweizer Infrastruktur

Tech Stack

Backend: FastAPI (Python), Uvicorn
AI: Ollama (qwen2.5:7b, qwen2.5vl:7b, granite3.2-vision)
Server: Infomaniak Swiss Cloud (NVIDIA L4, 24 GB VRAM)
TLS: Let's Encrypt (llm.poweron.swiss)

Deployment

Automatisches Deployment via Forgejo Actions bei Push zu main. Ziel: /opt/ollama-webapp/app/ auf 83.228.200.109.