# route-switch

> An OpenAI-compatible LLM gateway from Skelf Research. Routes requests across
> multiple providers (OpenAI, Anthropic, Google, Ollama, Cohere, Mistral via gollm),
> tracks per-prompt analytics in DuckDB, and reruns MIPROv2 prompt optimization
> against captured production traces. Written in Go. MIT licensed. Self-hosted.

## Project

- Repository: https://github.com/Skelf-Research/route-switch
- Documentation: https://docs.skelfresearch.com/route-switch/
- Organization: Skelf Research (https://www.skelfresearch.com)
- License: MIT
- Language: Go 1.24+

## Core pages

- [Home](https://route-switch.skelfresearch.com/): overview, what-is answer, problem→solution, features, code showcase, providers.
- [Features](https://route-switch.skelfresearch.com/features/): full capability surface — gateway, routing, optimization, operations.
- [How it works](https://route-switch.skelfresearch.com/how-it-works/): request lifecycle + architecture diagram.
- [Quickstart](https://route-switch.skelfresearch.com/quickstart/): build → config → run → route (HowTo).
- [FAQ](https://route-switch.skelfresearch.com/faq/): general, routing, optimization, operations.
- [Glossary](https://route-switch.skelfresearch.com/glossary/): gateway, combination, strategy, MIPROv2, goptuna, gollm, DuckDB, trace.
- [About](https://route-switch.skelfresearch.com/about/): why it exists, what's in the box, what it deliberately isn't.
- [Contact](https://route-switch.skelfresearch.com/contact/): reach the team.

## Guides

- [Guides index](https://route-switch.skelfresearch.com/guides/): all how-to guides for running the gateway and optimizer.
- [Drop-in OpenAI base_url swap](https://route-switch.skelfresearch.com/guides/drop-in-openai-base-url-swap/): point an existing OpenAI client at the gateway with a base_url change.
- [Choosing a routing strategy](https://route-switch.skelfresearch.com/guides/choosing-a-routing-strategy/): pick a strategy and weights for quality, cost, or latency.
- [Capturing traces to DuckDB](https://route-switch.skelfresearch.com/guides/capturing-traces-to-duckdb/): log production requests as traces for later optimization.
- [Running MIPROv2 on production traces](https://route-switch.skelfresearch.com/guides/running-miprov2-on-traces/): rerun MIPROv2 against captured traffic to tune prompts.

## Use cases

- [Use cases index](https://route-switch.skelfresearch.com/use-cases/): worked scenarios the gateway is built for.
- [Multi-provider failover](https://route-switch.skelfresearch.com/use-cases/multi-provider-failover/): fall back across providers when one degrades or errors.
- [Cost-aware routing](https://route-switch.skelfresearch.com/use-cases/cost-aware-routing/): route to cheaper combinations without losing quality floors.
- [Latency-aware routing](https://route-switch.skelfresearch.com/use-cases/latency-aware-routing/): bias routing toward faster provider/model combinations.
- [Prompt optimization from production traces](https://route-switch.skelfresearch.com/use-cases/prompt-optimization-from-traces/): close the loop from captured traffic to optimized prompts.
- [OpenAI-compatible migration](https://route-switch.skelfresearch.com/use-cases/openai-compatible-migration/): move off a single OpenAI dependency with no client rewrite.

## Comparisons

- [Comparisons index](https://route-switch.skelfresearch.com/compare/): how route-switch differs from adjacent tools.
- [vs OpenRouter](https://route-switch.skelfresearch.com/compare/openrouter/): hosted aggregator vs self-hosted gateway.
- [vs Portkey](https://route-switch.skelfresearch.com/compare/portkey/): observability-first vs optimization-first.

## Notes

- [Notes index](https://route-switch.skelfresearch.com/blog/): grounded essays on routing, prompt optimization, and the trade-offs.
- [Multi-model routing as a feature, not a hack](https://route-switch.skelfresearch.com/blog/multi-model-routing-as-a-feature/): why a gateway beats an in-app if-tree around provider SDKs.
- [Per-model prompt optimization: what actually moves the score](https://route-switch.skelfresearch.com/blog/per-model-prompt-optimization-what-moves-the-score/): what MIPROv2 and the evaluation strategies actually improve.
- [Quality / cost / latency: the routing triangle](https://route-switch.skelfresearch.com/blog/quality-cost-latency-the-routing-triangle/): the serving trilemma and three config postures for it.

## Feeds

- [RSS](https://route-switch.skelfresearch.com/rss.xml)
- [Sitemap](https://route-switch.skelfresearch.com/sitemap-index.xml)
- [Full LLM dump](https://route-switch.skelfresearch.com/llms-full.txt)