An agent needs an activity trail
A coding agent makes a sequence of decisions: what context to inspect, which tool to request, how to interpret the result, and whether a change is complete. Observability makes that sequence visible enough for a developer to review and correct.
The goal is not to expose every hidden model thought. The useful record is operational: active provider and model, selected mode, context sources, tool requests, tool results, errors, cancellation, changed files, and validation output.
What to inspect when a run goes wrong
Start with the concrete failure. Did the provider reject a key, rate-limit a request, return malformed tool arguments, or lose the relevant file context? Those are different problems and they need different fixes. A generic 'agent failed' message hides the evidence needed to solve any of them.
- Confirm the selected provider, model, and account.
- Inspect the tool that failed and its validated arguments.
- Check the returned error before retrying.
- Review the diff and validation results after a write.
Visibility improves trust and iteration
When tool activity is visible, developers can interrupt a bad path early, tighten the task, or switch to a more appropriate model. It also makes a successful run easier to reproduce because the important actions are not hidden behind one final message.
Truss shows tool activity alongside conversations and keeps Git-aware review close to the working session, so an agent run can be treated like normal engineering work rather than a mysterious output generator.
Next steps
Put it to work in your own workspace.
Pick the surface that fits your day, then connect a model you control.
FAQ
Questions developers ask before switching.
What is agent observability?
It is the ability to inspect the operational behavior of an agent, including context selection, tool requests, results, failures, and final changes.
Do coding agents need logs?
They need useful activity records. The record should help a developer diagnose provider, tool, context, or validation problems without exposing unnecessary sensitive content.