Merge remote-tracking branch 'origin/farm/d3ff0cb8/detect-lm-studio-vision'

packages/coding-agent/CHANGELOG.md

@@ -65,6 +65,7 @@
 - Fixed `/plan <prompt>` and `/goal <objective>` to preserve the typed slash-command line in TUI input history when entering those modes from off ([#2887](https://github.com/can1357/oh-my-pi/issues/2887)).
 - Fixed `/model` in the TUI to open the active-session model switcher instead of the role-assignment picker ([#2846](https://github.com/can1357/oh-my-pi/issues/2846)).
 - Fixed Perplexity web search collapsing every upstream failure to a generic `401 No authentication method available` once all auth methods failed: the fallback loop now rethrows the last classified provider error (`402`/credits-exhausted, `429`, `5xx`), so quota and rate-limit failures are no longer mis-reported as authorization errors. The generic 401 is now only a defensive fallback for the no-method-ran case.
+- Fixed LM Studio runtime discovery to use native `/api/v0/models` metadata so `inspect_image` can select VLM models. ([#2945](https://github.com/can1357/oh-my-pi/issues/2945))
 
 ### Security
 

packages/coding-agent/src/config/model-discovery.ts

@@ -13,6 +13,7 @@ import {
 	resolveModelReference,
 	stripBracketedModelIdAffixes,
 } from "@oh-my-pi/pi-catalog/identity";
+import { fetchLmStudioNativeModelMetadata } from "@oh-my-pi/pi-catalog/provider-models/openai-compat";
 import type { ModelSpec } from "@oh-my-pi/pi-catalog/types";
 import { isRecord } from "@oh-my-pi/pi-utils";
 import type { ProviderDiscovery } from "./models-config-schema";
@@ -379,18 +380,25 @@ export async function discoverOpenAIModelsList(
 	const baseHeaders: Record<string, string> = { ...(providerConfig.headers ?? {}) };
 	let headers = baseHeaders;
 	const attempt = async (h: Record<string, string>) => {
-		const res = await ctx.fetch(modelsUrl, {
-			headers: h,
-			signal: AbortSignal.timeout(10_000),
-		});
+		const nativeMetadataPromise =
+			providerConfig.discovery.type === "lm-studio"
+				? fetchLmStudioNativeModelMetadata(baseUrl, ctx.fetch, { headers: h })
+				: Promise.resolve(null);
+		const [res, nativeMetadata] = await Promise.all([
+			ctx.fetch(modelsUrl, {
+				headers: h,
+				signal: AbortSignal.timeout(10_000),
+			}),
+			nativeMetadataPromise,
+		]);
 		if (!res.ok) {
 			throw new Error(`HTTP ${res.status} from ${modelsUrl}`);
 		}
 		headers = h;
-		return res;
+		return [res, nativeMetadata] as const;
 	};
 	const apiKey = await ctx.getBearerApiKeyResolver(providerConfig.provider);
-	const response = apiKey
+	const [response, nativeMetadata] = apiKey
 		? await withAuth(apiKey, key => attempt({ ...baseHeaders, Authorization: `Bearer ${key}` }))
 		: await attempt(baseHeaders);
 	const payload = (await response.json()) as {
@@ -401,8 +409,12 @@ export async function discoverOpenAIModelsList(
 	for (const item of models) {
 		const id = item.id;
 		if (!id) continue;
+		const nativeMetadataForModel = nativeMetadata?.get(id);
 		const contextWindow =
-			toPositiveNumberOrUndefined(item.max_model_len) ?? toPositiveNumberOrUndefined(item.context_length) ?? 128000;
+			toPositiveNumberOrUndefined(item.max_model_len) ??
+			toPositiveNumberOrUndefined(item.context_length) ??
+			nativeMetadataForModel?.contextWindow ??
+			128000;
 		discovered.push(
 			buildModel({
 				id,
@@ -411,7 +423,7 @@ export async function discoverOpenAIModelsList(
 				provider: providerConfig.provider,
 				baseUrl,
 				reasoning: false,
-				input: ["text"],
+				input: nativeMetadataForModel?.input ?? ["text"],
 				cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
 				contextWindow,
 				maxTokens: Math.min(contextWindow, discoveryDefaultMaxTokens(providerConfig.api)),

packages/coding-agent/test/lm-studio-fix.test.ts

@@ -60,6 +60,53 @@ describe("ModelRegistry LM Studio Fixes", () => {
 		expect(available.some(m => m.provider === "lm-studio")).toBe(true);
 	});
 
+	test("marks LM Studio native VLM models as image-capable", async () => {
+		const fetchMock: FetchImpl = input => {
+			const url = String(input);
+			if (url === "http://127.0.0.1:1234/api/v0/models") {
+				return Promise.resolve(
+					new Response(
+						JSON.stringify({
+							data: [
+								{
+									id: "qwen/qwen3.6-27b",
+									type: "vlm",
+									capabilities: ["tool_use"],
+									max_context_length: 262144,
+								},
+								{ id: "plain-llm", type: "llm" },
+							],
+						}),
+						{ status: 200, headers: { "Content-Type": "application/json" } },
+					),
+				);
+			}
+			if (url === "http://127.0.0.1:1234/v1/models") {
+				return Promise.resolve(
+					new Response(
+						JSON.stringify({
+							data: [
+								{ id: "qwen/qwen3.6-27b", object: "model" },
+								{ id: "plain-llm", object: "model" },
+							],
+						}),
+						{ status: 200, headers: { "Content-Type": "application/json" } },
+					),
+				);
+			}
+			return Promise.resolve(new Response(null, { status: 404 }));
+		};
+
+		const registry = new ModelRegistry(authStorage, modelsJsonPath, { fetch: fetchMock });
+		await registry.refresh();
+
+		const vision = registry.find("lm-studio", "qwen/qwen3.6-27b");
+		const text = registry.find("lm-studio", "plain-llm");
+		expect(vision?.input).toEqual(["text", "image"]);
+		expect(vision?.contextWindow).toBe(262144);
+		expect(text?.input).toEqual(["text"]);
+	});
+
 	test("LM_STUDIO_BASE_URL can target any local OpenAI-compatible /v1 server", async () => {
 		const originalBaseUrl = Bun.env.LM_STUDIO_BASE_URL;
 		Bun.env.LM_STUDIO_BASE_URL = "http://127.0.0.1:11434/v1";