[DevMando]

Summary

Desktop users can now discover and pull Ollama models without leaving the app. Type /ollama-pull or choose Browse models… from the model dropdown to open a native model browser.

For example, a user can find a model, compare its variants and sizes, see which exact tags are already available on their server, pull a new variant, and explicitly choose Use in this agent when ready.

What changed

• Browse the library: search model families, sort by Newest, Popular, or Name, and inspect descriptions, capabilities, tags, and sizes. Reuse the CLI's catalog parser and streaming pull service.
• Recognize existing models: check the originating agent's Ollama server and show exact-tag Pulled or Available on server badges. Installing one variant does not mark its siblings installed. An unavailable inventory is shown as unknown rather than an empty list.
• Make selection easier: selecting a family automatically chooses its default tag. Cloud-only browsing shows cloud variants, preserves choices when a family has several cloud tags, and counts only installed cloud variants. Sorting or changing the cloud filter clears previous details and rejects late responses from older requests.
• Pull with feedback: show streamed status and stage progress, support cancellation and retry, and allow an exact model tag when the public catalog is unavailable. Refresh server inventory after a successful or cancelled pull.
• Keep agent scope clear: the separate, non-modal window identifies its originating agent and server. Pulling does not change the agent's model. Use in this agent refreshes availability, respects the agent's busy state, and refuses a switch if its endpoint changed. Closing the browser or originating agent cancels browser requests.
• Match Desktop: use native WinUI controls, the current app theme, display-scaling-aware sizing, and the same MandoCode window icon as the main app.
• Update the bundled engine: pin the published MandoCode CLI v0.16.2 release (1087b8a), replacing v0.16.1. This includes the latest CLI startup/focus rendering fix.
• Update the README, Unreleased changelog, and source-folder guide.

Why this was done

The CLI already offered model discovery and pulls, while Desktop required users to manage downloads elsewhere. This brings that workflow into Desktop and adds the missing installed-model visibility so users can choose the right variant with less guesswork.

How it was tested

• Final Release Desktop build passed with zero warnings and errors against the newly pinned CLI v0.16.2 engine.
• All 652 Desktop tests passed in Release, including 21 new cases covering exact tag identity, cloud filtering, default tag selection, stale lookup rejection, inventory failures, streamed pull responses, cancellation, and disposal.
• Validated the shared parser against the live Ollama catalog: 269 models and 58 Qwen3 tags returned during validation.
• Interactive checks covered opening from /ollama-pull and the model dropdown, search, exact installed-variant badges, automatic cloud-tag selection with enabled actions, clearing details when switching sort, theme appearance, and display scaling. The project owner also tested and refined the workflow.
• Pull transport was exercised with simulated streamed server responses. No model downloads were started during automated validation. The final cloud-filter refinements and window icon were compiled and covered where applicable by unit tests; they were not rechecked interactively in the final pinned-engine build.
• Staged patch whitespace checks passed.

Notes

• Desktop's application version remains unchanged in this feature PR. After merge, the next release is planned as Desktop v0.16.0 with CLI v0.16.2 bundled.
• The local engine launchSettings.json edit is excluded. No CLI source changes or unpublished engine branch commits are included.
• Installed badges indicate availability, not whether a newer model revision exists. Users can explicitly choose Pull again to refresh a model.