Closed Beta

Understand why your AI agent fails

AgentSpan sits between your agent and the LLM provider, captures every request and shows exactly what went wrong — in a readable interface for the whole team. Two config lines, no SDK.

Failure rate alerts

Get notified when your agent starts failing more than usual. Catch regressions before your users do.

No SDK required

Works out of the box — no SDK, just change two config lines. Cloud or self-host, your choice.

Near-zero overhead

~59ms per request. Proxy is synchronous, analytics are async. Your agent barely notices.

Readable by non-engineers

Session timelines and failure clusters are readable without knowing prompts or token counts.

Works with your agents

AgentSpan integrates with any agent that uses a standard LLM SDK — just change the base URL.

Works today: OpenClaw, n8n, Cursor, Crew AI — via URL change.

Not supported: Claude Code (closed SDK), GitHub Copilot (closed SDK).

Two lines of code. Full visibility.

Change the base URL and API key — that's it. No SDK wrappers, no code changes beyond that.

1. Get your key

Create a free account and get your AgentSpan API key.

2. Point to the proxy

Change base_url to our proxy endpoint. Keep using your existing SDK.

3. Start observing

Every request is traced. Sessions and failures appear instantly.

Everything you need to debug agents

Built for engineers and product managers who want to build reliable AI products.

Session Timeline

Open any session and in 10 seconds understand exactly where the agent went wrong — without reading raw logs.

Failure Clusters

Discover that 60% of your errors share the same root cause. Fix one pattern — drop your failure rate at once.

Request Categories

See which tasks users actually give the agent — and which request types underperform.

Multi-provider

OpenAI, Anthropic, DeepSeek and any OpenAI-compatible API. Switch providers — traces stay in one place.

Pricing

Beta — free during testing

Free

1,000 spans/month, 3 members, 1 project, 30 days retention, basic narratives, deterministic clusters.

Pro

100,000 spans/month, 20 members, 10 projects, 90 days retention, LLM narratives, failure clusters, request categories, alerts.

Self-host

Unlimited spans, unlimited members, unlimited projects, unlimited retention, full source code, Docker Compose deploy.

Start in two minutes

No credit card. No SDK changes. Just change two lines of config.