Preload default Ollama model before suggestions

[FuJacob]

Summary

Preload Cotabby's default local Ollama model through its native API before the first autocomplete request, so cancellable typing work cannot abort the model's cold start. The preload has a 120-second timeout and keeps the model resident, while non-Ollama OpenAI-compatible endpoints remain untouched.

Validation

• swiftlint lint --quiet — exited 0.
• xcodebuild -project Cotabby.xcodeproj -scheme Cotabby -destination 'platform=macOS' build-for-testing -derivedDataPath build/DerivedData — TEST BUILD SUCCEEDED.
• xcodebuild -project Cotabby.xcodeproj -scheme Cotabby -destination 'platform=macOS' build -derivedDataPath build/DerivedData — BUILD SUCCEEDED.
• Focused OpenAICompatibleAPIClientTests execution could not load the app-hosted test bundle because the host and bundle have different Team IDs, the repository's documented local signing limitation. Compilation and linking of the test bundle succeeded.

Linked issues

None.

Risk / rollout notes

The Ollama-only preload is deliberately restricted to Cotabby's exact default endpoint (http://127.0.0.1:11434/v1). Custom OpenAI-compatible endpoints continue using only the standard API, and preload failures remain best-effort warnings rather than user-facing generation failures.

Greptile Summary

This PR adds native Ollama preloading for Cotabby's default local endpoint. The main changes are:

• A default-only /api/generate URL derived from the OpenAI-compatible base URL.
• A preload request that keeps the selected Ollama model resident.
• OpenAI-compatible engine prewarm logic that runs the preload best-effort.
• Tests for the default Ollama route and non-default endpoint no-op behavior.

Confidence Score: 5/5

This looks safe to merge after a small cleanup to avoid repeated local preloads.

• The native Ollama request is limited to the exact default loopback endpoint.
• The payload and timeout are covered by focused tests.
• Repeated focus changes can start duplicate preload requests for the same model.

Cotabby/Services/Runtime/OpenAICompatibleSuggestionEngine.swift

Important Files Changed

Filename	Overview
Cotabby/App/Coordinators/SuggestionCoordinator+Input.swift	Updates the prewarm comment to include the OpenAI-compatible Ollama preload path.
Cotabby/Models/OpenAICompatibleEndpointModels.swift	Adds a default-only native Ollama generate URL derived from the existing base URL.
Cotabby/Services/Runtime/OpenAICompatibleAPIClient.swift	Adds the native Ollama preload request and status validation.
Cotabby/Services/Runtime/OpenAICompatibleSuggestionEngine.swift	Adds best-effort preload behavior, with a duplicate-request cleanup noted for repeated focus changes.
CotabbyTests/OpenAICompatibleAPIClientTests.swift	Adds coverage for Ollama URL derivation, preload payload fields, timeout, and non-default endpoint behavior.

_{Reviews (1): Last reviewed commit: "Preload default Ollama model before sugg..." | Re-trigger Greptile}

Greptile also left 1 inline comment on this PR.

[greptile-apps]

Duplicate Ollama Preloads

When focus moves across editable fields while the default Ollama model is still loading, each prewarm call starts another 120-second /api/generate request for the same model. These tasks are not tracked or cancelled by the prediction work controller, so rapid focus changes can queue duplicated local model-load work before the first autocomplete request.