Respan
Respan is an LLM engineering platform for teams shipping language-model and agent products in production. Route requests through one gateway, capture every call as a trace, run evaluations on live traffic, and manage prompts from a shared dashboard. It is built for engineering and product teams who need one place to debug, measure, and improve AI features after launch.
The product grew out of Keywords AI, which started as an LLM routing API at the University of Illinois before expanding into full observability. Respan now unifies gateway routing, tracing, evals, and prompt management around a single span data model, so production logs feed directly into quality scoring and iteration.
Platform teams, AI engineers, and startups running agents at scale use Respan to watch cost and latency, catch regressions, and ship prompt or model changes without redeploying application code. Customers cited on the site include Retell AI, Lovable, Gumloop, and Mem0.
Route OpenAI-style calls to 500+ models through one gateway
Automatic fallback and retries when a model errors or rate-limits
Every gateway call becomes a trace tree with latency on each span
Compose evaluators with LLM judges, code checks, and human review
Set soft warnings or hard spend caps per API key with Slack alerts
Free tier includes the full platform with 100k logs and no credit card required.
Gateway supports 500+ models with fallbacks, retries, caching, and spend limits in one place.
Combines tracing, evals, prompt management, and gateway routing on shared span data.
SOC 2, GDPR, ISO 27001, and HIPAA compliance options are documented on the site.
Team plan includes only five member seats before per additional member.
Free tier log retention is seven days compared with 30 days on Team.
Gateway adds roughly 50 to 150ms latency when routing through Respan instead of direct provider calls.
Does Respan have a free plan?
Yes. Respan offers a Free plan at with the full platform, 100k logs, 1k scores, 5 datasets, 2 evaluators, and 5 prompts. No credit card is required to sign up.
What models does Respan support?
Respan routes requests to 500+ models through its AI gateway. You can send OpenAI-style calls through Respan or keep each provider native SDK on a passthrough endpoint while every request is logged.
What integrations does Respan support?
Respan integrates with frameworks and tools including LangChain, LlamaIndex, Vercel AI SDK, OpenAI SDK, Mastra, Mem0, PostHog, and LiteLLM. Python and JavaScript SDKs are documented on respan.ai/docs.
Is Respan HIPAA compliant?
Respan states it is HIPAA compliant and offers a Business Associate Agreement for healthcare organizations on Enterprise plans. A HIPAA compliance add-on is listed at per month on the pricing page.
What was Respan formerly called?
Respan was previously known as Keywords AI. The company rebranded to Respan in February 2026 while expanding from LLM routing into a full observability and evaluation platform.

