Use Kortyx Studio
Updated 19 days ago • October 18, 2026
Kortyx Studio shows what happened when your application ran an agent. Follow a run from its first input through model calls, tools, pauses, and the final result.
Your application runs the agent. Studio receives its activity and lets you inspect it. You can use managed Cloud or run Studio yourself.
Choose your path
- Explore the sample project to learn the interface before connecting an app.
- Get started with Cloud for a managed installation.
- Run Studio locally to try it on your machine.
- Connect your application if Studio is already set up.
- Explore your first run if your application is already sending data.
- Deploy Studio yourself to operate a team installation.
Use Studio
Start by opening one run. Once you can follow its steps, choose the question you need to answer:
- Why did it fail? Inspect the trace and errors.
- Why was it slow or expensive? Investigate performance.
- What happened across several messages? Follow a session.
- Will a change improve the answer? Run an evaluation.
- How do I change instructions safely? Create and test a prompt.
What the system contains
The SDK application executes your workflows. A telemetry API stores the activity, and Studio displays it. Cloud operates the API, database, and Studio for you; self-hosting puts those services on your infrastructure. The core concepts explain how runs, sessions, and workflow maps fit together.
What Studio captures
Studio can show workflow steps, model timing, token counts, costs, and errors. Your application chooses what to send. Prompt text and inputs/outputs are excluded by default; content capture requires an explicit application setting.
Child workflow visibility
A workflow can call another workflow and use its result. Studio shows these child calls inside the parent execution. Inspect child workflows for waiting, returned, inherited, or reused work.
Evaluate workflow behavior
Evaluations run example conversations against your application and check the results. They execute your real agent through an authenticated endpoint. Start with your first evaluation, then read and compare results.