Chat with files and folders
Ask about selected files and verify the AI’s results.
UpdatedOn this page
Choose context
Open the file or folder chat tool from the available tools for that item. Confirm the path and select a configured model. Cloud models may receive file content relevant to the task; check for sensitive information first.
Ask a specific question
Start with a clearly scoped request, for example:
Summarize the three main conclusions in this document and identify the section supporting each one.
For folders, start by asking about the directory structure, then focus on a particular file. Do not assume every subfolder and large file has been fully read. Check tool calls and results to understand the actual scope.
Verify the answer
AI results can be inaccurate. Verify important numbers, citations, and file changes against the original. Read any operation prompts before continuing with tool execution.
For network or model errors, first try a simple text-only conversation, then check provider configuration.
Agent plugins and skills
Chat uses the bundled dsh runtime. In Chat settings, manage Agent plugins, MCP servers, and skills. Plugins can be installed, enabled, disabled, or removed; inspect the expandable logs if installation fails. Windows MCP is available on Windows and requires its managed Python environment. Browser and computer operations may request conversation-specific approval, which can be revoked in plugin settings. Import skills from Markdown or ZIP. Existing chat history is migrated into separate versioned storage; if recovery is requested, choose the offered database or backup. External CLI, MCP, and A2A setup is covered in Agent integration.
Capture and annotate screenshots
Use Capture in Chat or ⌥⌘4 on macOS / Ctrl+Alt+4 on Windows; add Shift to hide the current window. Windows also supports PrintScreen and Shift+PrintScreen. Select a region or window, add shapes, arrows, text or mosaic, and select annotations to move, resize or delete them. Drag the toolbar by its handle or empty space. Complete saves a PNG in Downloads and copies it to the clipboard; a capture started from Chat is also attached to that conversation. You can paste an image into Chat. On macOS, grant screen recording access when prompted.
Something unclear or different in the app? Let us know.