Feature Request / PR: Add configurable Connection Timeout for LM Studio provider to prevent 5-minute disconnections

[linusky]

Description:
Currently, when using LM Studio as the API Provider for tasks that require long generation times, the connection gets silently dropped after exactly 5 minutes. LM Studio logs show [LM STUDIO SERVER] Client disconnected. Stopping generation....

This happens because VS Code's native global fetch patches the connection and enforces a strict 300-second timeout, ignoring custom dispatcher settings. To fix this, we need to bypass the global fetch by using undici directly and expose a configurable timeout setting in the UI for LM Studio users.

Proposed Solution & Implementation Details:
I have successfully implemented and tested this locally by making the following changes:

1. State Keys (src/shared/storage/state-keys.ts):
Added lmStudioTimeoutMs to API_HANDLER_SETTINGS_FIELDS:

const API_HANDLER_SETTINGS_FIELDS = {
	// ... existing fields
	lmStudioMaxTokens: { default: undefined as string | undefined },
	lmStudioTimeoutMs: { default: undefined as number | undefined },
	// ...
}

2. LM Studio Provider (src/core/api/providers/lmstudio.ts):
Updated LmStudioHandlerOptions to include lmStudioTimeoutMs?: number.
Modified the createOpenAIClient call to explicitly use undiciFetch with a custom Agent to ensure the timeout limits are respected:

import { Agent, fetch as undiciFetch } from "undici"
import { createOpenAIClient } from "@/shared/net" // Removed globalFetch

// ... inside ensureClient()
const timeoutMs = this.options.lmStudioTimeoutMs ?? 20 * 60 * 1000 // Default to 20 mins
this.client = createOpenAIClient({
	baseURL: new URL("api/v0", this.options.lmStudioBaseUrl || "http://localhost:1234").toString(),
	apiKey: "noop",
	timeout: timeoutMs,
	fetch: (input, init) => {
		return undiciFetch(input as any, {
			...init,
			dispatcher: new Agent({
				headersTimeout: timeoutMs,
				bodyTimeout: timeoutMs,
				connectTimeout: timeoutMs,
			}),
		} as any) as any
	}
})

3. Settings UI (webview-ui/src/components/settings/providers/LMStudioProvider.tsx):
Added a new UI field for users to configure the timeout value in the LM Studio settings panel:

<div className="font-semibold">Connection Timeout</div>
<DebouncedTextField
	initialValue={apiConfiguration?.lmStudioTimeoutMs?.toString() || ""}
	onChange={(value) => {
		const numValue = parseInt(value, 10)
		if (!isNaN(numValue) && numValue > 0) {
			handleFieldChange("lmStudioTimeoutMs", numValue)
		} else if (value === "") {
			handleFieldChange("lmStudioTimeoutMs", undefined as any)
		}
	}}
	placeholder="Default: 1200000 (20 minutes in ms)"
	style={{ width: "100%" }}
/>

Why this is important:
Local LLM inference often requires much more time than cloud APIs, especially on older machines or with very large context windows. Giving users control over the undici fetch timeout ensures they can use Cline with local LM Studio instances without arbitrary 5-minute cutoffs.

Would love to see this integrated into the main branch!
Thank you

[linusky]

Hello,
Do my request is too much ? Because I use the version 3.84 (patched with the feature mentionned above) and it works really well with LM Studio, no more issue for long works.
Any chance to see this feature added to the next versions of Cline ?

Olivier

[asage-me]

I need this timeout setting as well