The hosted service: keys, dashboard and self-hosting

With an API key, the same spans go to AgentOps' dashboard, where traces are listed, drawn as trees and timelines, and costs added up. It can also be self-hosted.

import agentops

agentops.init(api_key="YOUR_AGENTOPS_KEY", default_tags=["production"])

Remove the endpoints, pass the key or set AGENTOPS_API_KEY, and init exchanges the key for a token that the exporter adds to every request. The course never does this: each run would send your traces to an account, and the dashboard's views could not be checked here.

• Session list: traces, filterable by tags and end state.
• Tree and timeline: the spans you printed with tree, with durations.
• Messages: LLM prompts and completions, unless content tracing is off.
• Costs: token costs from LLM spans plus tool costs.

Keeping data in your own infrastructure

AgentOps' backend is open source and documented for Docker. With a self-hosted backend, endpoint and exporter_endpoint point at your servers, which is the same change this course makes. Because the SDK speaks OpenTelemetry, exporter_endpoint can also point at any OTLP collector your customer already runs.