What it does

The official site at keywordsai.co shows the product under the name Respan. It describes an LLM engineering platform in which model requests pass through a single gateway, calls can be inspected as traces, spend shows up in metrics, and evaluations can be run. The gateway lets you keep the SDK you have and change its endpoint to reach over 1,000 models across major providers, with automatic fallbacks when a model errors or is rate-limited, response caching, and spend budgets and request limits that can be scoped to a key, a customer or the whole organization. Evals can combine an LLM judge, code checks and human review into one evaluator, draw test cases from production logs or CSV uploads, compare model or prompt versions on a dataset, and score live production spans. Monitoring puts request volume, error rate, cost, latency and token counts on one dashboard, with alerts to Slack, email or a webhook. Tracing records each model call, tool execution, retrieval step and agent turn as a span that carries the step's inputs, outputs, timing and cost. The site lists ISO 27001, SOC 2, GDPR and HIPAA compliance, with a BAA available.

Capabilities

One gateway endpoint to 1,000+ models, with fallbacks and response caching
Spend budgets and request caps that can be set for a key, a customer or the whole organization
Evals using LLM judges, code checks or human review, on test sets or live traffic
Tracing of model calls, tool executions, retrieval steps and agent turns
Monitoring dashboard with alerts to Slack, email or a webhook

When to use it

You call several LLM providers and want one endpoint with automatic fallbacks
You need to debug agent runs by inspecting what each step received, returned, took and cost
You want to score prompt or model changes before shipping them
You want cost and error alerts on production LLM traffic