Interact with the agent

Agent Session View overview

The Agent Session View holds one conversation with an agent. Here you can find your messages and the agent's replies, add context, configure the agent's settings, and perform other actions.

1. Run mode: Choose where the agent runs:

   • Local – on your machine, in the project directory.

   • Temporary Worktree – on your machine, in a separate Git working copy of the project.

   • Cloud – on JetBrains infrastructure, in a cloud clone of your repository.

2. Project: Select the project that the session works with.

3. Prompt library: Save prompts and reuse them for recurring tasks instead of typing them again.

4. Add: Attach project files, folders, commits, or other context to the request.

5. Start Voice Input: Dictate your request instead of typing it. Air in IDEs turns the recorded speech into prompt text.

6. Send: Send the request to the agent.

7. Agent preset: Click the selected agent to choose another agent preset.

8. Agent settings: Configure the model, effort, and operation mode for the next request. The availability of these settings depends on the agent.

9. Authorization: Select the authorization source that the agent works through. For more information, refer to Select the authorization for a session.

After you send the first prompt, a context indicator also appears in the text field.

Hovering over it shows how much of the model's context window is used in the current session. Prompts, attached context, and agent responses contribute to this usage, and the indicator updates as the session continues.

Configure agent settings

The chosen preset starts the agent with pre-configured settings, such as the model, effort level, and operation mode. You can change these settings for the current session using two controls below the text field:

• Model and effort settings: To configure the model and its settings, either click the model name or press Ctrl+Shift+M and select the required options:

  

  • Model – click to open the list of available models and select the model that processes your requests.

  • Effort – use the slider to select how much reasoning effort the model uses.

  • Other model-related options – enable or disable other model settings, such as Fast mode.

• Mode: Choose which actions the agent can perform without asking for confirmation. The popup can also hold agent modes, such as Plan or Goal.

  

Reduce AI credit usage with Eco mode in Junie

Junie, JetBrains' coding agent, provides an Eco mode that can reduce AI credit usage by sending routine implementation work to a smaller, cheaper companion model while keeping your selected model for planning and more complex tasks. Junie selects the companion model automatically, so you do not need to choose it separately.

Eco mode does not replace the model that you select. Junie uses your selected model to plan the task and make complex decisions. After planning, it routes suitable routine implementation steps to the companion model. The work of both models appears in the same session, as one conversation with the agent. The companion model can differ depending on the selected model, its provider, and the effort level.

Eco mode is not available for every model. Whether you can turn it on depends on the selected model, its provider, and the effort level. The free Lite model of Junie Lite does not support Eco mode.

To turn Eco mode on or off, use the Eco mode option in the model and effort settings. When Eco mode is on, appears next to the model and effort settings below the text field.

If you select Default in the Junie model list, Eco mode is turned on automatically. This option offers a recommended balance between the quality of the results and the cost.

Attach context

Click Add in the text field to search for context to add. The popup lists the recent files first and offers the following sources:

• Project Files and Folders: Any file or directory in the project.

• Upload from Computer: A file outside the project, such as a log or a specification.

• Context from Screen: A screenshot. On macOS 14 and newer, this opens the native window and display picker, which can capture another application even when its window is behind the IDE, and shows an Attach/Retake/Cancel preview before you add the capture.

• Commits: One or more commits from the repository history.

• Skills: A skill that tells the agent how to perform a specific kind of task.

Air in IDEs adds these attachments to the prompt that the agent receives, so the agent works with the context required for the task.

As an alternative, you can use the Add to Agent Context action to hand a specific part of your project to an agent. To do this:

Compose and send a request

To prepare a prompt for the agent, write your request in the text field using natural language. If you need to span the request across multiple lines, press Shift+Enter to start a new line.

Alternatively, you can dictate the request, reuse your previous prompts, use suggestions provided by Air in IDEs when composing a prompt, accept follow-up prompts suggested by the agent, or reply to a part of an agent response.

Dictate a prompt

Instead of typing a request, you can dictate it. Click Start Voice Input in the prompt field, dictate the request, and click to finish the recording.

Air in IDEs transcribes the recording and adds it to the prompt.

Use prompt suggestions

After you add context, Air in IDEs can show predefined prompt suggestions in the text field. The suggestions are based on the type and context of the attachments and help you formulate a request for them.

Click a suggestion to add it to the request, then edit as needed before sending it.

Use follow-up prompts

After completing a task or a step, some agents can suggest a follow-up prompt for the next possible action. For example, an agent can offer to commit the changes after editing files, or to delete the files that it found safe to remove. The suggestion appears in the empty text field.

Press Tab to insert the suggested prompt into the text field, then edit it as needed and send it to the agent.

Reply to a part of an agent response

To ask the agent about a specific part of its response, select the text in the response and click Reply next to the selection.

Air in IDEs adds the selected text to the text field as a quote. The request is not sent yet, so add your question or comment after the quote, then send the request to the agent.

Reuse prompts

To reuse your previous prompts, click the Open prompt library button in the top-right corner of the text field and select the required entry from the Prompts dialog.

You can save a prompt by clicking the Save to Prompts button next to it. Saved prompts appear at the beginning of the list.

Add a prompt to a queue

While the agent is working, you can draft another prompt in the text field and add it to the queue. The agent processes queued prompts in order after it finishes the current request.

To manage a queued prompt, use the controls displayed next to it:

• Click Send now to send the prompt immediately.

• Click the button and select Edit Message to change the prompt, Delete to remove it, or Open in New Chat to start a new session with the selected prompt.

  Select Turn Off Queueing to stop sending the agent queued prompts, pausing their execution.

• Hold a queued prompt by the icon and drag it to the required position in the queue.

If the queue is paused, click Resume next to the queue header to continue processing the queued prompts.

Answer agent questions

Sometimes an agent might request specific details from you. In these cases, Air in IDEs displays an interactive form containing a question and a list of suggestions on how to proceed with the task. The session pauses until you submit or dismiss the form, and your answer is added to the chat as a single summary block.

Use the up and down arrow keys to navigate the choices. Press Enter to confirm the selected choice, or press Esc to close the form.