55 lines
1.2 KiB
Markdown
55 lines
1.2 KiB
Markdown
# Oghma RAG Proxy
|
|
|
|
RAG (Retrieval-Augmented Generation) proxy for SkyrimNet that injects Tamrielic lore into NPC conversations based on their knowledge profile.
|
|
|
|
## Overview
|
|
|
|
This proxy sits between SkyrimNet and your LLM inference endpoint (OpenRouter/vLLM), enriching prompts with relevant lore from CHIM's Oghma Infinium database.
|
|
|
|
**Key Features:**
|
|
- Zero changes to SkyrimNet - just change the endpoint URL
|
|
- NPC-aware filtering - guards don't know mage secrets
|
|
- Two-tier knowledge - scholars get deep lore, commoners get basics
|
|
- ChromaDB-powered semantic search
|
|
|
|
## Quick Start
|
|
|
|
```bash
|
|
# Install
|
|
pip install -e .
|
|
|
|
# Ingest Oghma lore into ChromaDB
|
|
python -m oghma_proxy.ingest --host iris-dev.eachpath.local --port 35000
|
|
|
|
# Run proxy
|
|
python -m oghma_proxy.main
|
|
```
|
|
|
|
## Configuration
|
|
|
|
Copy `config.yaml` to `config.local.yaml` and customize:
|
|
|
|
```yaml
|
|
upstream:
|
|
url: https://openrouter.ai/api/v1
|
|
api_key: ${OPENROUTER_API_KEY}
|
|
|
|
chromadb:
|
|
host: iris-dev.eachpath.local
|
|
port: 35000
|
|
```
|
|
|
|
## Kubernetes Deployment
|
|
|
|
```bash
|
|
kubectl apply -k k8s/
|
|
```
|
|
|
|
## Architecture
|
|
|
|
See [TECHNICAL-SPEC.md](TECHNICAL-SPEC.md) for full design documentation.
|
|
|
|
---
|
|
|
|
Part of the [nimmerverse](https://github.com/dafit/nimmerverse) project.
|