Observability for AI agents

Every agent decision,
visible in one place.

Send structured agent events. Find failed steps, slow calls, and costly runs before your users report them.

1,000 runs per month. No card required. Node.js SDK is MIT-licensed.

agent timeline
RUN-1847guarded
LLM Requestgpt-4o · prompt 2,340 tok
LLM Response1,247 tokens · 1.73s
Tool Calldb.query({ user_id: '***' })
Guardrailpolicy check · reviewed
CompleteRun finished successfully
Tokens3,587
Cost$0.004
Latency1.92s
VerdictGuarded

Get started in minutes

import { Guardonic } from "@guardonic/sdk"
const g = new Guardonic({ apiKey, baseUrl })
await g.ingest({ agent, events })

The Node.js SDK is MIT-licensed. The hosted API and dashboard are proprietary; integrate from Python or other languages via our HTTP API. Node SDK documentation.

How it works

Instrument, observe, and control AI agents in production

Instrument

Add the Node SDK or call the HTTP API. Send the model, tool, status, timing, token, and cost data your team needs.

Observe

Open a run timeline and move from a high-level failure signal to the exact model or tool event that needs attention.

Improve

Compare success, latency, token use, and cost across agents. Use the evidence to fix prompts, tools, and workflows.

Every AI agent in production is a black box. Most teams find out something went wrong from their customers.

01

An agent calls an API it was never supposed to touch. You hear about it from a customer ticket.

02

A workflow burns through your monthly budget in six hours. No one set a limit.

03

Compliance asks what your agent told a user last Tuesday. There are no logs.

Platform

Everything you need to trust your agents in production.

Agent timeline

Every run broken down step by step. What the agent saw, decided, and did.

Simulation sandbox

Coming soon

Replay past runs with different inputs. Test before you ship.

Guardrails engine

In development

Cost caps, allowlists, and content policies enforced while a run is active.

Alerts

Coming soon

Slack, email, webhook. Know in seconds when something goes wrong.

Node SDKOpen source (MIT)
Run timelinesModel and tool events
InsightsLatency, tokens, and cost
REST APIAny language
VersionedTelemetry contract

Start monitoring, or talk to us first.

Create a workspace now, or tell us about your agents and what visibility you need.

We usually reply within one business day.