fix: improve local model provider robustness and UX

[Stijnus]

Summary

• Eliminates duplicated Docker URL rewriting code (was copy-pasted 4 times across Ollama/LMStudio)
• Adds error handling and 5-second timeouts to all local provider model fetches
• Fixes Ollama provider mutating internal SDK state by using createOllama() properly
• Fixes LLMManager singleton ignoring env updates on subsequent Cloudflare Worker requests
• Adds connection status indicators (green/red dots) for local providers in the model selector
• Shows helpful "is X running?" messages when a local provider has no models

Issues Fixed

1. Duplicated code: Docker URL rewriting was identical in 4 places — extracted to BaseProvider.resolveDockerUrl()
2. No error handling: Ollama/LMStudio getDynamicModels had zero try/catch — a crashed server would throw unhandled TypeError: fetch failed
3. No timeouts: One slow/hung local provider blocked all providers via Promise.all — added 5s AbortSignal.timeout
4. Fragile SDK usage: Ollama mutated ollamaInstance.config.baseURL via type cast — replaced with createOllama({ baseURL })
5. Stale env in singleton: LLMManager.getInstance(env) ignored env after first call — now updates on each call
6. Cache key too broad: Serialized entire serverEnv into cache key — narrowed to only provider-relevant keys
7. as any casts: LMStudio/OpenAILike used serverEnv as any — replaced with shared convertEnvToRecord()
8. No connection feedback: Users had no idea if local providers were running — added status dots and helpful empty-state messages

Files Changed

• app/lib/modules/llm/base-provider.ts — Added resolveDockerUrl(), convertEnvToRecord(), createTimeoutSignal(), narrowed cache key
• app/lib/modules/llm/providers/ollama.ts — Rewrote with shared helpers, proper createOllama(), error handling
• app/lib/modules/llm/providers/lmstudio.ts — Rewrote with shared helpers, error handling
• app/lib/modules/llm/providers/openai-like.ts — Added timeout, replaced as any, used logger
• app/lib/modules/llm/manager.ts — Fixed singleton env update, fixed typo
• app/components/chat/ModelSelector.tsx — Added local provider status dots and helpful empty states

Test plan

• Select Ollama provider with Ollama running — green dot, models listed
• Select Ollama provider with Ollama stopped — red dot, "is Ollama running?" message
• Select LMStudio provider with LM Studio stopped — red dot, helpful message
• Verify model loading doesn't hang when one local provider is unreachable (5s timeout)
• Verify cloud providers (OpenAI, Anthropic, etc.) are unaffected

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR improves the robustness and user experience of local model providers (Ollama, LMStudio, OpenAILike) by addressing code duplication, adding error handling and timeouts, fixing SDK usage patterns, and enhancing the UI with connection status indicators.

Changes:

• Extracted duplicated Docker URL rewriting logic to shared BaseProvider.resolveDockerUrl() method
• Added 5-second timeouts and comprehensive error handling to local provider model fetches
• Fixed Ollama provider to use createOllama() properly instead of mutating internal SDK state
• Fixed LLMManager singleton to update env bindings on subsequent calls (important for Cloudflare Workers)
• Added connection status indicators (green/red dots) and helpful error messages in the model selector UI

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
app/lib/modules/llm/base-provider.ts	Added shared helper methods (resolveDockerUrl, convertEnvToRecord, createTimeoutSignal) and narrowed cache key to include only provider-relevant env keys
app/lib/modules/llm/providers/ollama.ts	Refactored to use shared helpers, replaced SDK mutation with proper createOllama() usage, added error handling and timeout for model fetching
app/lib/modules/llm/providers/lmstudio.ts	Refactored to use shared helpers, added error handling and timeout for model fetching, removed as any cast
app/lib/modules/llm/providers/openai-like.ts	Added timeout to model fetch, replaced as any casts with proper type conversion, switched from console.log to logger utilities
app/lib/modules/llm/manager.ts	Fixed singleton to update env on subsequent calls, fixed typo in variable name (filteredStaticModesl → filteredStaticModels)
app/components/chat/ModelSelector.tsx	Added connection status indicators for local providers and helpful empty-state messages when providers are unreachable

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The chained .replace() calls only replace the first occurrence of each string. If a URL contains multiple instances of 'localhost' or '127.0.0.1' (unlikely but possible in some edge cases like 'http://localhost:1234/localhost'), only the first occurrence will be replaced. Consider using .replaceAll() or a regex with the global flag (e.g., .replace(/localhost/g, 'host.docker.internal')) to ensure all occurrences are replaced.

Suggested change

	return baseUrl.replace('localhost', 'host.docker.internal').replace('127.0.0.1', 'host.docker.internal');
	return baseUrl
	.replace(/localhost/g, 'host.docker.internal')
	.replace(/127\.0\.0\.1/g, 'host.docker.internal');

[Copilot]

The call to _resolveBaseUrl is outside the try-catch block. If _resolveBaseUrl throws an error (e.g., "No baseUrl found for Ollama provider"), it will not be caught and will propagate as an unhandled exception. Consider moving this line inside the try block to ensure all errors are properly handled and an empty array is returned.

[Copilot]

The call to _resolveBaseUrl is outside the try-catch block. If _resolveBaseUrl throws an error (e.g., "No baseUrl found for LMStudio provider"), it will not be caught and will propagate as an unhandled exception. Consider moving this line inside the try block to ensure all errors are properly handled and an empty array is returned.

Suggested change

	const baseUrl = this._resolveBaseUrl(apiKeys, settings, serverEnv);
	try {
	let baseUrl: string;
	try {
	baseUrl = this._resolveBaseUrl(apiKeys, settings, serverEnv);

[Copilot]

The call to _resolveBaseUrl can throw an error (e.g., "No baseUrl found for LMStudio provider"), but there's no error handling in getModelInstance. If this throws, it will propagate as an unhandled exception. Consider adding try-catch error handling or documenting that this method can throw.

Suggested change

	const baseUrl = this._resolveBaseUrl(apiKeys, providerSettings?.[this.name], envRecord);
	let baseUrl: string;
	try {
	baseUrl = this._resolveBaseUrl(apiKeys, providerSettings?.[this.name], envRecord);
	} catch (error) {
	logger.error('Error resolving LMStudio base URL in getModelInstance:', error);
	if (error instanceof Error) {
	throw error;
	}
	throw new Error('Error resolving LMStudio base URL in getModelInstance');
	}

[Copilot]

The call to _resolveBaseUrl can throw an error (e.g., "No baseUrl found for Ollama provider"), but there's no error handling in getModelInstance. If this throws, it will propagate as an unhandled exception. Consider adding try-catch error handling or documenting that this method can throw.

Suggested change

	const baseUrl = this._resolveBaseUrl(apiKeys, providerSettings?.[this.name], envRecord);
	let baseUrl: string;
	try {
	baseUrl = this._resolveBaseUrl(apiKeys, providerSettings?.[this.name], envRecord);
	} catch (error) {
	logger.error('Failed to resolve Ollama base URL', error);
	throw error;
	}