LiteRegistry Docs¶
Usage-focused guides for running and operating LiteRegistry. Start with the main README for a full end-to-end workflow, then use these pages for argument details and behavior.
| Guide | What it covers |
|---|---|
| CLI | Full literegistry subcommand reference and Fire usage |
| Registry | Redis vs filesystem backends, how entries are stored, inspecting the roster |
| Gateway | OpenAI-compatible proxy, CLI args, endpoints, retries |
| vLLM & SGLang | Launching model servers, registry registration, passthrough flags |
| Code & Terminal | Python executor and restricted log pipelines |
| Load balancing | Exp3 bandit routing, failover, latency feedback |
| Runtimes | local vs apptainer, binds, env, image pull |
| Console | Streamlit dashboard for gateway / vLLM / registry |
Typical flow¶
literegistry redis → registry backend
literegistry vllm / sglang → model workers (heartbeat into registry)
literegistry code / terminal→ tool workers (optional)
literegistry gateway → single HTTP front door
literegistry console → live ops view (optional)
All CLI commands go through literegistry <subcommand> (Fire). Registry URLs
are either redis://host:port or a filesystem path.
This site is built with MkDocs and published via GitHub Pages from master.