Integration guide

Connect your AI systems.
See what actually happens at runtime.

Your applications send metadata about what happened: which agent ran, which model it used, which tools were invoked, which databases and APIs were reached, how long it took, and whether errors occurred. ObserveAgents turns that evidence into an AI inventory, a dependency map, a capability map, and findings.

Try our Demo → Watch the product tour
OpenTelemetry collects what happened.
ObserveAgents explains what it means.
How to connect

From instrumentation
to runtime intelligence.

1

Install AI runtime instrumentation

Add OpenTelemetry-compatible instrumentation such as OpenLLMetry, OpenLit, OpenInference, or manual OTel spans.

2

Export traces through OTLP

Send traces to ObserveAgents using the standard OTLP endpoint.

3

Discover agents automatically

ObserveAgents identifies running agents from service.name, gen_ai attributes, tool calls, MCP calls, providers, models, and runtime behavior.

4

Build runtime intelligence

The platform maps each agent's tools, MCP usage, APIs, databases, providers, errors, owners, and environments.

5

Detect risky behavior

Detection Rules identify patterns such as high MCP usage, repeated tool errors, unknown providers in production, DB/API access, and missing ownership.

6

Recommend control

High-risk agents are surfaced in Gateway Control Center for review and optional control planning.

Ready to connect your first agent?

Try the live demo first — runtime timelines, the inventory, findings, and the Gateway Control Center on realistic synthetic data. Then connect your own OTel stack in an afternoon.

Try our Demo →
The ObserveAgents console