Connect your agent
See your models at work
Start your Python agent with automatic capture. Model calls, tokens, timing, errors, and supported framework activity appear here as you use it.
OpenAIAnthropicLangChain / LangGraphOpenAI-compatible local models
Default capture includes model names, token usage, timing, and errors. Enable this to also store message content locally.
1 Install in your agent’s environment
Download the installer, then run this command from your download folder in the Python environment your agent uses.
2Start your agent with automatic capture
This receiver requires a token. Set OBSERVATORY_TOKEN in your agent’s terminal before running the command.
Receiver listening
Waiting for your first agent
Connect an agent above, then make a model call. This page updates automatically.
Usage trends
Token use and run outcomes over time
Agent activity
Select a trace
Spans, progress and model usage appear here.
—
—
—
—
—
Metric explorer
Select an OTLP metric
Recent metric points
OTLP metric points will appear here.
Select a run to inspect its work.
Model interactions
Opt-in captured prompts and outputs will appear here.
Progress
Progress events linked to this trace will appear here.
Agent activity ledger
Structured agent actions linked to this trace will appear here.
Correlated logs
Logs linked to this trace will appear here.
Run context
Service and model details will appear here.
Related traces
Findings in this trace
Recorded failures and review findings will appear here.
Evaluation outcome
No outcome has been joined to this trace.