Local agents in the desktop app

Besides the cloud AI, the chat panel in the desktop app can use coding agents that run on your computer. They work in workspaces connected to a local folder: they read and write the files in that folder directly, with a full shell, Python and Node.js, and the app syncs their changes up in the background.

Pick an agent

Choose one in the agent selector at the bottom of the chat panel:

When the workspace isn't connected to a local folder, the local agents are dimmed; click "Connect to a local folder…" to connect one.

Installing and signing in to Claude Code and Codex

When the selector shows "Not installed", expand it to see the official install commands (ready to copy) and the install docs. After installing, click "Check again".

Before first use, run claude or codex once in a terminal and sign in when prompted. VizChat uses the copy you installed and your sign-in; it doesn't store your password or key separately.

The built-in Pi and models

Pi's model choice is the same as the cloud AI's on the web: official models use VizChat credits, and keys you add under Settings → Custom Model use your own account. A custom endpoint ("Other (OpenAI-compatible)") can be any address, including a model server on your company network or LAN — the desktop app connects to it directly, without going through VizChat's servers.

Offline, official models and keys relayed through the server are unavailable; a custom endpoint on your computer or LAN keeps working.

What the agent can do

The agent comes with VizChat's skills and the vizchat command line: it can read and write boards and widgets, check and render them (on your computer, without pushing first, also offline), and search the web. When it wants to change files or run commands, it asks according to the agent's own permission settings; the confirmation appears in the chat.

Closing the panel doesn't interrupt a turn in progress: the agent finishes, and the result is in the conversation. Quitting the app stops every turn in progress.

Conversation history

Conversations with local agents are kept on your computer and synced to your account: you can see them on other devices and on the web, but only continue them on the computer they started on. When a conversation can't be continued (the agent's own record is gone, or it came from elsewhere), use "Summarize and continue in a new conversation" — the summary is written by the local agent you picked.

The same command in your terminal

On macOS, choose "VizChat → Install vizchat Command…" from the menu. After that, vizchat in any terminal is the copy that comes with the app: it uses the account signed in to the app, and connected folders are synced by the app. See the local workspace command line.