Keywords AI (Respan)

Keywords AI began as a cost-conscious LLM API and monitoring layer for AI startups, and it now presents itself as Respan, an open-source gateway. The core idea is that a team building with language models should not have to choose between cheap routing and deep visibility. You point an existing OpenAI, Anthropic or Google SDK at the gateway's base URL, swap in one API key, and requests are routed, retried, load-balanced and logged without rewriting application code.

The observability half of the product records each call as a trace, tracks cost and usage per model, and supports prompt management with versioning and one-click deployment. Evaluations, datasets and LLM-as-judge scoring let teams run structured experiments and regression tests as prompts evolve, which is the part most ad-hoc logging setups lack.

Operationally the gateway claims very high throughput and low added latency, with fallbacks when a provider fails, per-customer spend limits and PII masking before data reaches a model. A free workspace covers a moderate log volume; paid team and enterprise tiers raise the limits and add collaboration, SSO and compliance options. Because it is open source, self-hosting is possible for organisations with stricter data requirements.