What it does
Respan is an AI gateway for applications that call language models. You keep your existing SDK, change its base URL and key to Respan's endpoint, and name the model in each request so it is routed there. The site says this reaches more than 1,000 models across the major providers, and it also offers a passthrough URL for people who want to keep a provider's native SDK. The gateway adds reliability and cost controls. You list fallback models and Respan moves down the list when a request errors or hits a rate limit. Response caching serves repeat requests without another upstream call, and budgets and rate limits can be set per key, per customer or across an organisation, with warnings as usage approaches a limit and blocking once it passes. Every request is logged. You can attach a customer identifier and custom properties such as feature or environment, then filter logs, traces and cost figures by them. The site describes this built-in observability and evals as what sets the gateway apart. To start, you can top up credits or bring your own provider keys. The site lists ISO 27001, SOC 2, GDPR and HIPAA compliance, with a BAA available for healthcare teams.
Capabilities
- One endpoint for 1,000+ models, switched by changing the model name
- Automatic fallback to listed backup models on errors or rate limits
- Response caching for repeated requests
- Budgets and rate limits per key, customer or organisation
- Request logs with customer identifiers and custom metadata
- Passthrough URLs for keeping a provider's native SDK
When to use it
- Switching between model providers without rewriting integration code
- Keeping an application running when one provider has an outage or rate limit
- Capping and tracking model spend by customer or feature
- Debugging model calls from one log of every request
