Skip to content

ai-setup.md

Latest commit

 

History

History
259 lines (172 loc) · 5.56 KB

ai-setup.md

File metadata and controls

259 lines (172 loc) · 5.56 KB
Raw

AI Setup

MeadowPy uses Ollama for local AI features. Ollama runs on your computer and serves models through a local HTTP API.

Related docs:

What MeadowPy Uses

MeadowPy talks to Ollama at this default API URL:

http://localhost:11434

The app checks:

  • GET / for server health.
  • GET /api/tags for installed model names.
  • POST /api/chat for streaming chat responses.

Chat responses stream into MeadowPy token by token.

Install Ollama

  1. Install Ollama from the official Ollama website.
  2. Start Ollama.
  3. Install at least one model.
  4. Open MeadowPy.
  5. Choose AI > Setup/check Ollama....
  6. Click Check Now.
  7. Select a model.
  8. Click Use Selected Model or Save Settings.

The setup dialog checks four things:

  • Whether the ollama command is available on PATH.
  • Whether the Ollama server is reachable.
  • Whether at least one model is installed.
  • Which model MeadowPy will use.

Install A Model

MeadowPy does not install models directly. Use Ollama to install a model, then return to MeadowPy and refresh the model list.

The general Ollama command shape is:

ollama pull <model-name>

Then verify that Ollama can see installed models:

ollama list

In MeadowPy, open AI > Setup/check Ollama... and click Check Now.

Select A Model

You can select a model in either place:

  • AI > Setup/check Ollama...
  • The Ollama/model label in the MeadowPy status bar.

When you select a model, MeadowPy saves it under:

ollama.selected_model

inside:

%USERPROFILE%\.meadowpy\settings.json

If the selected model is no longer returned by Ollama, MeadowPy clears the saved selection.

Auto Connect

By default, MeadowPy automatically checks Ollama on startup. The check repeats periodically while the app is open.

You can change this in:

File > Preferences > AI

or in:

AI > Setup/check Ollama...

The setting is:

ollama.auto_connect

AI Chat Panel

Open the AI Chat panel with:

AI > AI Chat Panel

or:

Ctrl+Shift+A

The chat panel can use editor context when available:

  • Current filename.
  • Current cursor line.
  • Enclosing function or class.
  • Current file text, unless the file is in large-file mode.

If no model is selected, MeadowPy reports:

No model selected. Click the AI status bar to choose one.

If Ollama is offline, MeadowPy reports:

Ollama is not connected.

Explain Selected Code

Select code in the editor, right-click, and choose Explain this code.

For Python files, MeadowPy sends a Python-specific explanation prompt. For other text files, it sends a general text explanation prompt.

Review And Improve Code

Select Python code, right-click, and choose Review & improve.

MeadowPy asks the AI to consider:

  • Readability.
  • Best practices.
  • Potential bugs.
  • Performance.
  • Structure.
  • Beginner-friendly explanations.

This action is only available for Python-mode editors.

Generate Docstrings

Right-click a Python function or class and choose Generate docstring.

MeadowPy asks the AI to return only a triple-quoted Python docstring inside a fenced code block. When you insert the result at the cursor, MeadowPy tries to place and indent the docstring correctly under the function or class.

Review Current File

Use:

AI > Review Current File

or:

Ctrl+Shift+R

For Python files, MeadowPy asks for feedback on:

  • Code structure and organization.
  • Readability and naming.
  • Potential bugs or issues.
  • Best practices and improvements.
  • Performance considerations.

For other text files, MeadowPy asks for feedback on clarity, organization, tone, wording, and consistency.

Full-file review is disabled for large-file mode. Select a smaller section and use Explain this code instead.

AI Analysis For Errors

When a Python program writes an error to stderr, the Output panel stores the last error text and shows AI Analysis.

Clicking AI Analysis sends:

  • The current editor text, if available.
  • The last error text.

The AI prompt asks what went wrong and how to fix it.

AI Analysis For Lint Issues

The Problems panel can request AI help for lint issues.

MeadowPy sends:

  • The linter code.
  • The linter message.
  • The affected line.
  • One nearby non-blank line above and below when available.

The AI prompt asks what the issue means and how to fix it.

Privacy Notes

MeadowPy's built-in AI integration is designed for a local Ollama server. With the default URL, prompts are sent to:

http://localhost:11434

That means MeadowPy sends prompt text to the local Ollama process on your machine, not directly to a cloud API.

Important limits:

  • Installed Ollama models are managed by the user.
  • MeadowPy cannot guarantee what a custom Ollama-compatible endpoint does.
  • If you change the API URL to a remote server, prompts may leave your machine.
  • AI responses can be wrong; review generated code before running it.

Troubleshooting AI

If AI features do not work:

  1. Open AI > Setup/check Ollama....
  2. Confirm the API URL is http://localhost:11434 unless you intentionally changed it.
  3. Click Check Now.
  4. If the command check fails, reinstall Ollama or reopen your terminal after installation.
  5. If the server check fails, start Ollama.
  6. If the model list is empty, install a model with Ollama.
  7. If a model is installed but not selected, choose it and save settings.

See Troubleshooting for more detailed fixes.