Respan

Respan (formerly Keywords AI) is an LLMOps platform built for teams shipping AI agents in production. Founded by Andy Li, Raymond Huang, and Hendrix Liu — who met at the University of Illinois — the company joined Y Combinator's Winter 2024 batch and has since rebuilt around a core insight: developers need deep visibility into how agents behave, not just routing. The platform covers the full loop from tracing and evaluation to prompt optimization, deployment, and monitoring.

What It Is

Respan is an AI observability and evaluation platform designed specifically for agent workflows. It captures every prompt, tool call, and model response in production, surfaces quality and performance signals automatically, and gives teams the controls to iterate without losing track of what changed. The platform positions itself as an alternative to tools like LangSmith, Langfuse, and Braintrust, with a unified system that combines tracing, evals, prompt management, and an AI gateway in one product.

How the Five-Stage Workflow Fits Together

Respan organizes its product around five connected stages:

• Trace — End-to-end execution paths capture every step from input to output, with search, filter, and sort by content, latency, cost, quality, tags, and custom metadata. Production traces can be replayed in a playground to debug failures in full context.
• Evaluate — Evaluation workflows combine human review, code checks, and LLM judges in a single flow. Teams define metrics first, then treat each judge as a function inside one evaluation system.
• Optimize — Prompt, tool, model, and workflow changes are versioned so teams always know what changed and when. New versions are tested against prior baselines using the same production data.
• Deploy — A single AI gateway routes across 500+ models with version control, rollout logic, and provider abstraction. Prompt and workflow versions can be promoted to production directly from the UI.
• Monitor — Custom dashboards with 80+ graph types track quality, latency, cost, and product-specific signals. Alerts fire via Slack, email, or text when behavior drifts; automations can trigger dataset builds or follow-up evaluations from production signals.

Integration Breadth

Respan supports a wide range of frameworks and providers out of the box:

• Frameworks: LangChain, LlamaIndex, Vercel AI SDK, Mastra, Haystack, BAML, Instructor, Agno, RubyLLM, OpenAI Agents SDK
• Providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, AWS Bedrock, Groq, Fireworks, Together AI, Perplexity, OpenRouter, Mistral, DeepSeek, Cohere, xAI, and more
• Ecosystem tools: Mem0, Cognee, AssemblyAI, Linkup, PostHog, Zapier, Cursor, Slack

Python and JavaScript/TypeScript SDKs are available, and the platform supports OpenTelemetry for standardized instrumentation.

Compliance and Security Posture

Respan holds ISO 27001 certification and meets SOC 2, GDPR, and HIPAA requirements. A Business Associate Agreement is available for healthcare organizations. The platform offers cloud deployment by default, with self-hosted options available at the enterprise tier. PII masking, conditional log retention, and data retention management are included in higher tiers.

Background: From Keywords AI to Respan

The company announced its rebrand from Keywords AI to Respan in February 2026, describing the shift as a move from a routing-focused API toward a proactive observability platform built for the agent era. The about page states that the company started with an API to dynamically route LLM calls, then rebuilt around production observability after early customer feedback. The homepage claims the platform has processed 80 trillion+ tokens, attributing this figure to vendor-published marketing copy. The company is a 10-person team operating out of Alameda, California, and is backed by Gradient, Y Combinator, Hat-Trick Capital, and a range of angel investors from companies including Lovable, Retell AI, and Mem0.