What the agent can do

Letronna works through tools. Each one it uses shows in the chat, and anything that changes your computer asks first: see Asking first and modes.

Files

• It reads, searches, writes and edits files in the project folder, and shows a diff before each change.
• It reads Word, Excel, PowerPoint and PDF files as text, OpenDocument files, and EPUB books chapter by chapter. Search can look inside documents too.
• Scratch work, such as a script to reproduce a bug, can go in the system's temp folder, out of your project.
• With no project folder attached, a chat works in Letronna's own folder.

Commands

• It runs commands in the project folder, in bash where bash is installed.
• A command still running at its time limit carries on as a background job. /jobs lists them in the terminal, and the Jobs tab in the window shows them live.
• In the terminal, a line starting with ! runs as a command yourself, without asking, and the model sees what it printed with your next message:

!git status

Web search and pages

Letronna can read a web page and search the web. Switch search on with the Web switch in the window, or /search in the terminal.

By default, search goes through the gateway when it sells search (on the Letronna gateway, paid from your credit), and through DuckDuckGo when it cannot answer. The approval card says where a search goes and who pays. To use a search service of your own, open Settings > Search, or:

/search services

A chosen service is the only one used. To keep every search on DuckDuckGo, with no key and no cost:

/search keyless

Pictures

The model sees pictures you attach and pictures its tools open. In the window, drop or paste them into the box. In the terminal, Ctrl+V attaches the picture on the clipboard, dragging a picture file in attaches it, or:

/image screenshot.png

Files and folders by name

Type @ and a path to attach a file, a range of lines or a folder. The box offers paths as you type.

Controlling apps

On macOS, and on Linux under X11, Letronna can use desktop apps for you. It asks once for each app.

Speaking and listening

• Replies can be read aloud as they stream, paid from your credit. In the terminal, /voice turns it on or off.
• In the window, voice input turns what you say into text at the cursor.

Images, speech and video

Making images, speech and video, and turning audio into text, are covered in Hosted models and credit. A picture you attach can also be edited with edit_image, on the chat's own gateway.

Reply style

/style lists the reply styles and picks one; you can also write your own.

/style concise