Platform
If you are new to Introspection, Core concepts explains how a recipe, runtime, task, and conversation relate. The quickstart shows you how to run your first agent. The pages under Platform explain each part of the product.
Control Plane and Data Plane
Introspection separates platform settings from the data created when an agent runs. The Control Plane stores settings such as projects, credentials, and runtimes. The Data Plane runs tasks and stores their results. The separation determines where identity and credentials are stored, how tasks are isolated, and which infrastructure holds customer data.
| Plane | What it holds |
|---|---|
| Control Plane | Organizations, projects, deployments, API keys, integrations, recipes, runtimes, judges, experiments, endpoints, variables, and credentials |
| Data Plane | Tasks, conversations, events, judgements, observations, patterns, state, and artifacts |
The Data Plane can run in Introspection’s Managed Cloud, in your own cloud account through bring your own cloud (BYOC), or in an on-premises environment. See Deployment and infrastructure, Security, and Data handling for details about each option.
Improve an agent using production results
Production tasks create conversations and results. Feedback, observations, patterns, and product outcomes can help you find behavior to improve. They do not change the agent by themselves.
First, review the conversations related to a problem and decide what behavior you want to change. Then make the smallest recipe change that could improve the behavior. Use an offline evaluation to test the change against representative cases. Use a calibrated judge to check the change against a quality standard. Run an experiment when you need to compare versions using production traffic.
After you validate the change, review it and add it to the recipe. Deploy the new version, and use its production results to find the next improvement. The guides explain how to use each method.
Explore the platform
| Area | Start with |
|---|---|
| Account and access | Organizations and projects, Authentication and identity, Tags |
| Run agents | Runtimes, Runtime manifest, Operating runtimes, Bindings, Tasks and runs, Conversations, State and artifacts |
| Learn from production | Observations and patterns, Judges, Experiments |
| Integrations | GitHub, Connectors, Connectors reference, MCP and federation, LLM providers |
| Deploy and secure | Deployment and infrastructure, Security |
For details about stored data, residency, and retention, see Data handling.