Working with the agent
This section describes the Explyt chat, the agent modes, and the features that help you steer a task. If you would rather solve a concrete problem than study the interface, start with the practical guides.
Core features
| Feature | When to use it |
|---|---|
| Chat and attachments | Pass the agent a request, a file, a code selection, a log, or other context |
| Agent modes | Choose whether the agent should explore the project, change code, write tests, build a plan, debug, or review |
| Context, chats, and memory | Understand what applies in the current chat and what persists between chats |
| Subagents and parallel work | Split a large task between specialized agents |
| Project onboarding | Take the tutorial on AGENTS.md, Rules, and Skills for a new project |
| Message queue | Send your next instructions while the agent is still working |
| Chat compression | Continue a long task when the history approaches the context limit |
| Images | Pass a screenshot of the interface or of an error |
| Ask/Copy | Clarify or copy a selected fragment of a response |
| Simple mode | Hide the extra interface elements and keep the core actions |
How to choose a mode
- General suits most tasks and can call subagents.
- Ask explores the project without changing files.
- Code reads and edits code.
- Test specializes in tests.
- Plan prepares a sequence of steps and acceptance criteria.
- Debug checks hypotheses through the IDE debugger.
- Review performs a read-only check.
- Orchestrator distributes a large task across specialized subagents.
The full tool set per mode is listed in the feature matrix.
Practical tasks
- Understand unfamiliar code
- Change code safely
- Fix a bug
- Add tests
- Review the agent's changes
- Prepare Explyt to work with your project
Before a task that involves sensitive data, check permissions and safety, the Tools menu, and the edit scope.