Orange3-CanvasAgent 0.1.3

An Orange3 canvas agent widget with reversible AI commits.

Orange3 Canvas Agent

Orange3-CanvasAgent adds a new Orange widget named Canvas Agent. It lets you describe a workflow in plain language, previews the agent's structured canvas changes, and keeps those changes reversible until you choose Keep Changes or Revert AI Commit.

Version 0.1.3 focuses on install polish: a simpler Windows installer flow, clearer first-run guidance, and better repair steps when the widget is installed but not visible yet.

Release Links

• GitHub releases: https://github.com/pantagram1031/orange3-canvasagent/releases
• PyPI package: https://pypi.org/project/Orange3-CanvasAgent/
• Issue tracker: https://github.com/pantagram1031/orange3-canvasagent/issues

Install For Beginners On Windows

This is the recommended path if you already use the normal Orange desktop app and do not want to open a terminal.

1. Open the GitHub releases page.
2. Download CanvasAgentSetup.exe from the latest release.
3. Double-click the installer.
4. If Windows shows a security warning, verify the file came from the GitHub release page above, then continue only if you trust it.
5. In the installer window, wait for Orange detection to finish.
6. Select your Orange installation from the list.
7. If nothing is listed, click Choose Folder... and point to your Orange install folder.
8. Click Install / Repair.
9. Continue to Verify and run the smoke check.
10. On Finish, confirm the success message and click Open Orange.
11. When Orange launches, look in the widget toolbox for the Canvas Agent category.
12. Drag Canvas Agent onto the canvas.

If the widget still does not appear, jump to Troubleshooting.

Install For Coders

From PyPI

python -m pip install Orange3-CanvasAgent
python -m Orange.canvas --force-discovery

From Source

git clone https://github.com/pantagram1031/orange3-canvasagent.git
cd orange3-canvasagent
python -m pip install -e ".[dev]"
python -B -m unittest discover -v
python -m Orange.canvas --force-discovery

First Run

1. Open Orange.
2. In the left widget toolbox, find the Canvas Agent category.
3. Drag Canvas Agent onto the canvas.
4. Open the Setup tab.
5. Click Check Setup.
6. If needed, use Sign in to authenticate the installed codex CLI.
7. Click Test Agent.
8. Open the Chat tab and send a request such as:

Add a File widget and a Data Table widget, then connect them.

9. Review the proposed changes on the Preview tab.
10. Click Keep Changes to accept them or Revert AI Commit to undo them.

Troubleshooting

• Installer did not find Orange: Re-run the installer and use Choose Folder... to select the Orange installation directory.
• A black command window flashed: Install v0.1.3 or newer from the latest release. The wizard now hides subprocess windows during install, verification, and launch.
• Installed, but Canvas Agent is not visible in Orange: Close Orange, reopen it, and check the left widget toolbox for the Canvas Agent category. If you installed from PyPI or source, run python -m Orange.canvas --force-discovery once, then reopen Orange.
• The widget category is still missing: Re-run the Windows installer if you used CanvasAgentSetup.exe, or verify that your PyPI/source install went into the same Python environment used by Orange.
• Codex CLI missing: Install the codex CLI first, then return to the widget and click Check Setup again.
• Codex CLI signed out: Use the widget's Sign in button or run codex login in a terminal.
• Agent request failed: Open the Diagnostics tab and copy the report before filing an issue.

Screenshot Placeholders

• [Placeholder] Installer welcome and Orange detection screen
• [Placeholder] Installer finish screen after successful install
• [Placeholder] Orange toolbox showing the Canvas Agent category
• [Placeholder] Canvas Agent Setup tab
• [Placeholder] Canvas Agent Preview tab with Keep/Revert controls
• [Placeholder] Diagnostics tab with discovery report

What The Widget Does

• Adds a dedicated Canvas Agent widget category to Orange.
• Uses structured canvas actions instead of arbitrary model-generated Python.
• Keeps an in-memory checkpoint so each AI change can be accepted or reverted.
• Includes Setup, Chat, Preview, and Diagnostics tabs.
• Checks both Codex CLI readiness and Orange widget discovery state.