Respan

What is Respan?

Respan is an LLM engineering platform that gives AI teams a model gateway, production observability and automated evaluations in one place. Developers send model calls through Respan’s API or SDKs, and the platform routes them to 1,000+ models while logging every call, tool run and agent step. Teams then score outputs, manage prompt versions and set alerts on cost, errors and latency.

The company was previously called Keywords AI and rebranded as Respan in February 2026, according to EveryDev. Its Y Combinator profile lists the Winter 2024 batch, a 2023 founding, a San Francisco base and Andy Li as founder and CEO. Respan describes itself as self-driving observability for AI agents, and the platform was live in October 2026 with a free plan.

Key Features

• Unified AI gateway: Respan routes requests to 1,000+ models from providers such as OpenAI, Anthropic and Groq through one API. Automatic failover, caching, load balancing and per-key spend limits keep production apps running when a single provider slows or fails.
• Agent tracing: Respan records every LLM call, tool run and agent step as spans within a trace. Each trace shows latency, cost and token use, so engineers can replay a failed run and see exactly where an agent went wrong.
• Automated evaluations: Respan scores outputs with LLM judges, deterministic code checks or human reviewers. Teams can test models against datasets before release and sample live traffic to catch regressions after deployment.
• Prompt management: Respan versions prompts and workflows and compares them against a baseline before rollout. The free plan holds 5 prompts, while Team allows an unlimited number for larger product teams.
• Datasets from production: Respan turns sampled production traffic into test datasets and can generate synthetic test cases. This keeps evaluation sets close to what real users send rather than relying on hand-written examples.
• Dashboards and alerts: Respan’s dashboards chart requests, error rates, cost and tokens, sliceable by model or user. Alerts fire through Slack, email or webhook when cost, errors, latency or token use cross a threshold you set.
• Security and compliance: Respan states it holds ISO 27001 and SOC 2 and supports GDPR and HIPAA, with a BAA and DPA available. EveryDev also lists PII masking, data retention controls and SSO through Google or SAML.

Pros

• One platform instead of several: Respan combines gateway, tracing, evaluations and prompt management in one product. Teams that would otherwise run a router, an observability tool and an eval tool can cut vendors and integration work.
• Generous free tier: Respan’s free plan includes 100k logs, 1k scores, unlimited seats and 7-day retention. Small teams and solo developers can instrument a real app before deciding whether Team is worth paying for.
• Production scale claims: Respan says it has processed over 80 trillion tokens, and its Y Combinator profile reports 1B+ logs a month. Named customers include Retell AI, Mem0, Lovable and Gumloop, which suggests it handles serious traffic.
• Broad integrations: Respan offers Python and JavaScript SDKs and works with LangChain, LlamaIndex, the OpenAI SDK, Vercel AI and OpenTelemetry. Most stacks can start sending traces with a small code change.
• Enterprise-ready controls: Respan supports HIPAA BAAs, SSO, PII masking and custom SLAs on its higher tiers. Healthcare and regulated teams can meet procurement checks without bolting on separate tools.

Cons

• Steep jump to Team: Respan moves from a free plan straight to Team at $199 a month billed yearly. Small startups that outgrow the free limits have no low-cost middle tier to bridge the gap.
• Usage overage costs: Respan charges $8 per extra 100k logs and $1 per extra 1k scores beyond plan limits. High-traffic agents can push bills well above the headline Team price.
• Short free retention: Respan keeps logs for 7 days on the free plan and 30 days on Team by default. Teams that need longer audit trails will need Enterprise terms or their own export.
• Developer-only tool: Respan targets engineers who instrument code with SDKs, so non-technical users get little from it. Product managers can read dashboards, but setup and evaluations need developer time.
• Crowded, newer brand: Respan rebranded from Keywords AI in February 2026 and competes with LangSmith, Langfuse and Braintrust. Some older documentation and listings still use the previous name, which can cause confusion.

Who is Using Respan?

• AI startups: Trace agents, route models and control spend in production apps.
• LLM and platform engineers: Debug failed agent runs and replay traces with full context.
• ML evaluation teams: Build datasets from live traffic and score outputs with judges.
• Regulated enterprises: Use HIPAA BAAs, SSO and PII masking for compliant AI features.

Pricing

Respan has a free plan, a Team plan at $199 a month billed yearly and custom Enterprise pricing (official prices, October 2026).

Directories differ on model coverage, with FutureTools and EveryDev citing 500+ models against Respan’s own 1,000+. Usage-based overages can change your real monthly cost, so estimate log volume before choosing a tier.

What Makes Respan Unique?

A gateway and an evaluation suite sharing the same trace data is Respan’s main differentiator. Many rival tools focus on either routing or observability, so teams stitch two products together. Because Respan already sits in the request path, it can log, score and alert on every call without a second integration, then apply fixes through routing rules or prompt versions. That loop from trace to evaluation to deployment is what the company calls self-driving observability.

Compatibilities and Integrations

• SDKs and frameworks: Python and JavaScript SDKs, LangChain, LlamaIndex, OpenAI SDK and Vercel AI.
• Model providers: OpenAI, Anthropic, Groq, Ollama and other providers through one gateway.
• Standards: OpenTelemetry support for sending traces from existing instrumentation.
• Alerts and access: Slack, email and webhook alerts, plus Google and SAML single sign-on.

Respan Tutorials

Instrumenting a first app usually takes one short coding session.

1. Create a free account at respan.ai and generate an API key.
2. Install the Python or JavaScript SDK, or point your OpenAI client at the Respan gateway.
3. Run your app and open the traces view to inspect spans, cost and latency.
4. Build a dataset from sampled traffic and add an LLM judge or code check.
5. Set alerts for cost or error thresholds and route them to Slack or email.

How We Rated It

Summary

Respan suits engineering teams running LLM agents in production who want routing, tracing and evaluations in one platform. The main limitation is cost structure: the free plan is generous, but the next step is $199 a month plus usage overages. It also requires developer time to set up. Start on the free plan with one service, measure your monthly log volume, and only then compare Team against LangSmith or Langfuse.