selixes/gateway

Selixes is a self hosted layer that sits between your application and AI providers like OpenAI, Anthropic, or Google, aiming to keep your product working even when one of those providers has an outage or slows down. If your app currently talks directly to a single AI provider, you are depending on that provider's uptime. Selixes gives you a local control point instead. Setting it up is meant to take about five minutes. Since it is built to match the standard OpenAI SDK, you mostly just change two settings: your API key and the base URL, so requests go to your local Selixes instance instead of directly to OpenAI. From there you can add extra instructions to each request through HTTP headers, such as a timeout after which it should automatically switch to a backup provider like Anthropic, a maximum dollar amount a single session is allowed to spend before it gets blocked, and whether to use cached answers for similar prompts. A few features stand out in the README. It automatically detects failures like timeouts, rate limits, or connection errors and reroutes requests to a standby model. It uses vector based caching through Pinecone so that questions with similar meaning, not just identical text, can be answered instantly from cache instead of calling the AI provider again. During a complete outage, it can fall back to running a local model called Llama 3 through Ollama, so basic functionality continues at no extra cost. Because it runs inside your own private cloud, your prompts and data never pass through a third party service. To run it locally you need Docker and Node.js, and the project includes a dashboard for watching request activity and reviewing what happened when something failed. A separate guide covers deploying it to a real server for production use.