VorloVorlo in 60 seconds
Vorlo is the debugger for AI agents. Add two lines of code and every run of your agent becomes a replayable session — and when a step fails, Vorlo tells you the root cause in plain English and the exact fix.
How it works
- Trace— the SDK hooks into your agent's callbacks and captures every tool call, the model's reasoning, latency, and token cost. Sends are fire-and-forget: Vorlo can never slow down or crash your agent, even if our servers are unreachable.
- Diagnose — failures are translated from raw errors (
ToolError: 403) into a root cause and a specific fix. Common errors match a curated registry instantly; the long tail is diagnosed by an LLM with full step context. - Learn — every diagnosis carries a confidence tier (verified · likely · guess). When a fix is confirmed — by you, or automatically when the error stops recurring — it's promoted to verified for everyone who hits it next.
- Prevent — guardrail policies can block or flag risky actions before they run, based on failures Vorlo has already seen.
Pick your quickstart
Python →LangChain agents with
pip install vorlo-trace. Two lines to your first replay.JavaScript / TypeScript →LangChain.js in Node or Next.js with npm install vorlo-trace.What you get in the dashboard
- Step replay — a chronological walk through every step, with inputs, outputs, reasoning, and a latency waterfall. Running sessions update live.
- Failure diagnosis — root cause, fix, confidence tier, and one-click handoff (copy for your editor, your team chat, or an issue).
- Failure clusters — failures grouped by root cause across sessions, so you fix the pattern instead of the symptom.
- Tool reliability — per-tool success rate, p50/p99 latency, and live degraded-tool alerts.
Free during beta. Create an account at vorlo.dev/sign-up, generate an API key in Settings, and you're tracing in under three minutes.