Respan

What is Respan?

Respan is a production observability and evaluation platform built specifically for AI agents and LLM-powered applications. If you're shipping anything that calls a language model in production, whether that's a chatbot, a multi-step agent, or a document processing pipeline, this is the kind of tool that tells you what's actually happening inside it.

The platform captures full execution traces across messages, tool calls, routing decisions, memory reads, environment state, and final outcomes. Rather than seeing a single prompt and response, you get a timeline of every step an agent took, which makes debugging complex failures significantly faster. One user described the difference plainly: instead of spending an hour trying to reproduce a bug, they could see exactly what went wrong almost immediately.

Observability is only part of the picture. **Respan also handles evaluation**, letting you run automated scoring on production traffic or historical data. You can compare outputs across different model versions, prompt variants, or configurations and catch regressions before they affect users. The evaluation layer connects directly to the trace data, so findings aren't abstract metrics sitting in a separate dashboard.

A built-in LLM gateway lets you route traffic across more than a dozen providers including OpenAI, Anthropic, and others through a single base URL. Switching models or running A/B tests becomes a configuration change rather than a code change. Token usage, latency, and cost are tracked at the request level and rolled up by user, project, or environment, which is useful if you're billing clients or trying to understand where money is going.

Integration is reasonably painless. The OpenTelemetry-based SDK for Python and JavaScript uses decorators to instrument existing code without restructuring it. Teams that are already using LangChain or the Vercel AI SDK can connect with minimal changes. Respan processes over a billion logs and two trillion tokens monthly, so the infrastructure is clearly built to handle real scale.

The company was part of Y Combinator's W24 cohort, previously operating under the name Keywords AI before rebranding. It raised $5 million in seed funding from Gradient Ventures and Y Combinator in March 2026. The platform holds SOC 2, ISO 27001, and HIPAA certifications, which matters if you're building in regulated environments.

**Pricing starts free** with 100k logs and a limited evaluation quota. The paid tier for growing teams is a monthly subscription and unlocks unlimited datasets, evaluators, and prompt versions along with a private support channel. Enterprise pricing is custom.

For indie developers or small teams building AI products, the free tier is genuinely useful for getting started. The tool is developer-facing by nature, so there's no point pretending it's a no-code experience. If you're writing code that calls LLMs, Respan gives you the visibility you'd otherwise build yourself.