@oh-my-pi/pi-coding-agent 15.12.3 → 15.13.0

package/src/commands/say.ts

@@ -0,0 +1,102 @@
+/**
+ * Synthesize text with the local TTS engine and play it (or save it with --out).
+ *
+ * Demonstrates the on-device speech stack end to end: the first run downloads
+ * the configured local model, synthesis happens in the TTS worker subprocess,
+ * and the resulting WAV is either played through the speakers or written to disk.
+ */
+import * as os from "node:os";
+import * as path from "node:path";
+import { getProjectDir, Snowflake } from "@oh-my-pi/pi-utils";
+import { Args, Command, Flags } from "@oh-my-pi/pi-utils/cli";
+import chalk from "chalk";
+import { Settings, settings } from "../config/settings";
+import { playAudioFile, removeTempFile } from "../tts/player";
+import { shutdownTtsClient, ttsClient } from "../tts/tts-client";
+import { encodeWav } from "../tts/wav";
+
+export default class Say extends Command {
+	static description = "Synthesize text with the local TTS engine and play it through the speakers";
+
+	static args = {
+		text: Args.string({ required: true, description: "Text to speak" }),
+	};
+
+	static flags = {
+		voice: Flags.string({ description: "Voice id" }),
+		model: Flags.string({ description: "Local TTS model key" }),
+		out: Flags.string({ char: "o", description: "Write WAV to this path instead of playing" }),
+	};
+
+	static examples = [
+		'omp say "hello world"',
+		'omp say "hello world" --out /tmp/hello.wav',
+		'omp say "bonjour" --voice af_heart --model kokoro',
+	];
+
+	async run(): Promise<void> {
+		const { args, flags } = await this.parse(Say);
+		const text = args.text ?? "";
+
+		await Settings.init({ cwd: getProjectDir() });
+		const model = flags.model ?? settings.get("tts.localModel");
+		const voice = flags.voice ?? settings.get("tts.localVoice");
+
+		let exitCode = 0;
+		const unsubscribe = ttsClient.onProgress(event => {
+			if (event.status === "progress" && typeof event.progress === "number") {
+				process.stderr.write(
+					`\r${chalk.dim(`downloading ${event.file ?? model}: ${Math.round(event.progress)}%`)}`,
+				);
+			} else if (event.status === "done" || event.status === "ready") {
+				// Clear the progress line once the download finishes.
+				process.stderr.write("\r\x1b[K");
+			}
+		});
+
+		try {
+			const audio = await ttsClient.synthesize(model, text, { voice });
+			if (!audio) {
+				process.stderr.write(
+					chalk.red(
+						`error: could not synthesize with local TTS model "${model}". ` +
+							"Run `omp setup speech` to install it.\n",
+					),
+				);
+				exitCode = 1;
+				return;
+			}
+
+			const wav = encodeWav(audio.pcm, audio.sampleRate);
+			const durationSec = audio.pcm.length / audio.sampleRate;
+
+			if (flags.out) {
+				await Bun.write(flags.out, wav);
+				process.stdout.write(
+					`${chalk.green("saved")} ${flags.out} ` +
+						`${chalk.dim(`(${voice}, ${model}, ${durationSec.toFixed(1)}s, ${wav.byteLength} bytes)`)}\n`,
+				);
+				return;
+			}
+
+			const tmp = path.join(os.tmpdir(), `omp-say-${Snowflake.next()}.wav`);
+			await Bun.write(tmp, wav);
+			try {
+				await playAudioFile(tmp);
+				process.stdout.write(
+					`${chalk.green("spoke")} ${chalk.dim(`(${voice}, ${model}, ${durationSec.toFixed(1)}s)`)}\n`,
+				);
+			} finally {
+				await removeTempFile(tmp);
+			}
+		} catch (err) {
+			process.stderr.write(chalk.red(`error: ${err instanceof Error ? err.message : String(err)}\n`));
+			exitCode = 1;
+		} finally {
+			unsubscribe();
+			await shutdownTtsClient();
+		}
+
+		if (exitCode !== 0) process.exit(exitCode);
+	}
+}

package/src/commands/setup.ts

@@ -7,7 +7,7 @@ import { runSetupCommand, type SetupCommandArgs, type SetupComponent } from "../
 import { runRootCommand } from "../main";
 import { initTheme } from "../modes/theme/theme";
 
-const COMPONENTS: SetupComponent[] = ["python", "stt"];
+const COMPONENTS: SetupComponent[] = ["python", "speech"];
 
 export interface OnboardingSetupDependencies {
 	runRoot?: typeof runRootCommand;

package/src/commands/token.ts

@@ -0,0 +1,89 @@
+/**
+ * Get the API key or OAuth token for a provider.
+ */
+
+import { PROVIDER_REGISTRY } from "@oh-my-pi/pi-ai";
+import { Args, Command, Flags } from "@oh-my-pi/pi-utils/cli";
+import chalk from "chalk";
+import { isAuthenticated, ModelRegistry } from "../config/model-registry";
+import { discoverAuthStorage } from "../sdk";
+
+export default class Token extends Command {
+	static description = "Get the API key or OAuth token for a provider";
+
+	static args = {
+		provider: Args.string({
+			description: "Provider ID (e.g. anthropic, openai)",
+			required: true,
+		}),
+	};
+
+	static flags = {
+		raw: Flags.boolean({
+			description: "Output the raw credential value without parsing nested JSON structures",
+			default: false,
+		}),
+		"force-refresh": Flags.boolean({
+			description: "Force refresh the OAuth token even if it has not expired",
+			default: false,
+		}),
+	};
+
+	static examples = [
+		"# Get API key for Anthropic\n  omp token anthropic",
+		"# Get raw Copilot credential JSON\n  omp token github-copilot --raw",
+		"# Force refresh and get Gemini CLI token\n  omp token google-gemini-cli --force-refresh",
+	];
+
+	async run(): Promise<void> {
+		const { args, flags } = await this.parse(Token);
+		const providerName = args.provider ?? "";
+		const provider = providerName.toLowerCase();
+
+		const authStorage = await discoverAuthStorage();
+		const modelRegistry = new ModelRegistry(authStorage);
+
+		// Resolve the API key / token
+		const apiKey = await modelRegistry.getApiKeyForProvider(provider, undefined, {
+			forceRefresh: flags["force-refresh"],
+		});
+
+		if (!isAuthenticated(apiKey)) {
+			// Find all active/configured providers
+			const activeProviders = new Set<string>();
+			for (const p of PROVIDER_REGISTRY) {
+				if (authStorage.hasAuth(p.id)) {
+					activeProviders.add(p.id);
+				}
+			}
+			const all = authStorage.getAll();
+			for (const p in all) {
+				if (authStorage.hasAuth(p)) {
+					activeProviders.add(p);
+				}
+			}
+
+			const msg = `No active credential found for provider "${providerName}".`;
+			process.stderr.write(`${chalk.red(msg)}\n`);
+			if (activeProviders.size > 0) {
+				process.stderr.write(`Configured providers: ${Array.from(activeProviders).sort().join(", ")}\n`);
+			}
+			process.exitCode = 1;
+			return;
+		}
+
+		if (!flags.raw) {
+			try {
+				const parsed = JSON.parse(apiKey);
+				if (parsed && typeof parsed === "object" && typeof parsed.token === "string") {
+					process.stdout.write(`${parsed.token}\n`);
+					return;
+				}
+			} catch {
+				// Not a JSON string, print as-is
+			}
+		}
+
+		process.stdout.write(`${apiKey}\n`);
+	}
+}

package/src/commit/agentic/tools/analyze-file.ts

@@ -1,5 +1,5 @@
 import { prompt } from "@oh-my-pi/pi-utils";
-import * as z from "zod/v4";
+import { z } from "zod/v4";
 import analyzeFilePrompt from "../../../commit/agentic/prompts/analyze-file.md" with { type: "text" };
 import type { CommitAgentState } from "../../../commit/agentic/state";
 import type { NumstatEntry } from "../../../commit/types";
@@ -38,6 +38,9 @@ function buildToolSession(
 	return {
 		cwd: options.cwd,
 		hasUI: false,
+		// Programmatic fan-out: results feed the commit agent's evidence, not a
+		// model choosing further spawns, so the specialization nudge is noise here.
+		suppressSpawnAdvisory: true,
 		getSessionFile: () => ctx.sessionManager.getSessionFile() ?? null,
 		getSessionSpawns: () => options.spawns,
 		settings: options.settings,

package/src/commit/agentic/tools/git-file-diff.ts

@@ -1,4 +1,4 @@
-import * as z from "zod/v4";
+import { z } from "zod/v4";
 import type { CommitAgentState } from "../../../commit/agentic/state";
 import type { CustomTool } from "../../../extensibility/custom-tools/types";
 import * as git from "../../../utils/git";

package/src/commit/agentic/tools/git-hunk.ts

@@ -1,4 +1,4 @@
-import * as z from "zod/v4";
+import { z } from "zod/v4";
 import type { DiffHunk, FileHunks } from "../../../commit/types";
 import type { CustomTool } from "../../../extensibility/custom-tools/types";
 import * as git from "../../../utils/git";

package/src/commit/agentic/tools/git-overview.ts

@@ -1,4 +1,4 @@
-import * as z from "zod/v4";
+import { z } from "zod/v4";
 import type { CommitAgentState, GitOverviewSnapshot } from "../../../commit/agentic/state";
 import { extractScopeCandidates } from "../../../commit/analysis/scope";
 import type { CustomTool } from "../../../extensibility/custom-tools/types";

package/src/commit/agentic/tools/propose-changelog.ts

@@ -1,4 +1,4 @@
-import * as z from "zod/v4";
+import { z } from "zod/v4";
 import type { CommitAgentState } from "../../../commit/agentic/state";
 import { CHANGELOG_CATEGORIES, type ChangelogCategory } from "../../../commit/types";
 import type { CustomTool } from "../../../extensibility/custom-tools/types";

package/src/commit/agentic/tools/propose-commit.ts

@@ -1,4 +1,4 @@
-import * as z from "zod/v4";
+import { z } from "zod/v4";
 import type { CommitAgentState } from "../../../commit/agentic/state";
 import {
 	capDetails,

package/src/commit/agentic/tools/recent-commits.ts

@@ -1,4 +1,4 @@
-import * as z from "zod/v4";
+import { z } from "zod/v4";
 import type { CustomTool } from "../../../extensibility/custom-tools/types";
 import * as git from "../../../utils/git";
 

package/src/commit/agentic/tools/schemas.ts

@@ -1,4 +1,4 @@
-import * as z from "zod/v4";
+import { z } from "zod/v4";
 
 export const commitTypeSchema = z.enum([
 	"feat",

package/src/commit/agentic/tools/split-commit.ts

@@ -1,4 +1,4 @@
-import * as z from "zod/v4";
+import { z } from "zod/v4";
 import type { CommitAgentState, SplitCommitGroup, SplitCommitPlan } from "../../../commit/agentic/state";
 import { computeDependencyOrder } from "../../../commit/agentic/topo-sort";
 import {

package/src/commit/analysis/summary.ts

@@ -2,7 +2,7 @@ import type { ThinkingLevel } from "@oh-my-pi/pi-agent-core";
 import type { Api, ApiKey, AssistantMessage, Model } from "@oh-my-pi/pi-ai";
 import { completeSimple, validateToolCall } from "@oh-my-pi/pi-ai";
 import { prompt } from "@oh-my-pi/pi-utils";
-import * as z from "zod/v4";
+import { z } from "zod/v4";
 import summarySystemPrompt from "../../commit/prompts/summary-system.md" with { type: "text" };
 import summaryUserPrompt from "../../commit/prompts/summary-user.md" with { type: "text" };
 import type { CommitSummary } from "../../commit/types";

package/src/commit/changelog/generate.ts

@@ -2,7 +2,7 @@ import type { ThinkingLevel } from "@oh-my-pi/pi-agent-core";
 import type { Api, ApiKey, AssistantMessage, Model } from "@oh-my-pi/pi-ai";
 import { completeSimple, validateToolCall } from "@oh-my-pi/pi-ai";
 import { prompt } from "@oh-my-pi/pi-utils";
-import * as z from "zod/v4";
+import { z } from "zod/v4";
 import changelogSystemPrompt from "../../commit/prompts/changelog-system.md" with { type: "text" };
 import changelogUserPrompt from "../../commit/prompts/changelog-user.md" with { type: "text" };
 import { CHANGELOG_CATEGORIES, type ChangelogCategory, type ChangelogGenerationResult } from "../../commit/types";

package/src/commit/shared-llm.ts

@@ -1,6 +1,6 @@
 import type { AssistantMessage } from "@oh-my-pi/pi-ai";
 import { validateToolCall } from "@oh-my-pi/pi-ai";
-import * as z from "zod/v4";
+import { z } from "zod/v4";
 import type { ChangelogCategory, ConventionalAnalysis } from "./types";
 import { extractTextContent, extractToolCall, normalizeAnalysis, parseJsonPayload } from "./utils";
 

package/src/config/keybindings.ts

@@ -212,8 +212,8 @@ export const KEYBINDINGS = {
 		description: "Search history",
 	},
 	"app.stt.toggle": {
-		defaultKeys: "alt+h",
-		description: "Toggle speech-to-text",
+		defaultKeys: [],
+		description: "Toggle speech-to-text (default gesture: hold Space)",
 	},
 } as const satisfies KeybindingDefinitions;
 

package/src/config/model-discovery.ts

@@ -393,12 +393,16 @@ export async function discoverOpenAIModelsList(
 	const response = apiKey
 		? await withAuth(apiKey, key => attempt({ ...baseHeaders, Authorization: `Bearer ${key}` }))
 		: await attempt(baseHeaders);
-	const payload = (await response.json()) as { data?: Array<{ id: string }> };
+	const payload = (await response.json()) as {
+		data?: Array<{ id?: string; max_model_len?: unknown; context_length?: unknown }>;
+	};
 	const models = payload.data ?? [];
 	const discovered: Model<Api>[] = [];
 	for (const item of models) {
 		const id = item.id;
 		if (!id) continue;
+		const contextWindow =
+			toPositiveNumberOrUndefined(item.max_model_len) ?? toPositiveNumberOrUndefined(item.context_length) ?? 128000;
 		discovered.push(
 			buildModel({
 				id,
@@ -409,8 +413,8 @@ export async function discoverOpenAIModelsList(
 				reasoning: false,
 				input: ["text"],
 				cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
-				contextWindow: 128000,
-				maxTokens: discoveryDefaultMaxTokens(providerConfig.api),
+				contextWindow,
+				maxTokens: Math.min(contextWindow, discoveryDefaultMaxTokens(providerConfig.api)),
 				headers,
 				compat: {
 					supportsStore: false,
@@ -463,7 +467,7 @@ export async function discoverProxyModels(
 		? await withAuth(apiKey, key => attempt({ ...baseHeaders, Authorization: `Bearer ${key}` }))
 		: await attempt(baseHeaders);
 	const payload = (await response.json()) as {
-		data?: Array<{ id?: string; name?: string; supported_endpoint_types?: string[] }>;
+		data?: Array<{ id?: string; name?: string; supported_endpoint_types?: string[]; context_length?: number }>;
 	};
 	const items = payload.data ?? [];
 	const discovered: Model<Api>[] = [];
@@ -499,7 +503,9 @@ export async function discoverProxyModels(
 				// upstream bundled catalogs, so keep costs local-unknown even when
 				// we successfully recover the upstream model identity.
 				cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
-				contextWindow: reference?.contextWindow ?? 128000,
+				// Prefer the context_length the API reports for this model; fall
+				// back to the bundled reference, then a sane default.
+				contextWindow: toPositiveNumberOrUndefined(item.context_length) ?? reference?.contextWindow ?? 128000,
 				maxTokens: reference?.maxTokens ?? discoveryDefaultMaxTokens(api),
 				headers,
 				// OpenAI-compat fields are no-ops on anthropic models; the

package/src/config/model-registry.ts

@@ -17,8 +17,6 @@ import {
 	googleGeminiCliModelManagerOptions,
 	openaiCodexModelManagerOptions,
 	PROVIDER_DESCRIPTORS,
-	UNK_CONTEXT_WINDOW,
-	UNK_MAX_TOKENS,
 } from "@oh-my-pi/pi-catalog/provider-models";
 import {
 	collapseBuiltModelVariants,
@@ -26,9 +24,11 @@ import {
 	resolveVariantAlias,
 } from "@oh-my-pi/pi-catalog/variant-collapse";
 
-// Sentinel for local-only OAuth token (LM Studio, vLLM) — declared inline to avoid loading
-// any provider module at startup. Must match `DEFAULT_LOCAL_TOKEN` in oauth/lm-studio.ts.
+// Sentinels for local-only OAuth tokens — declared inline to avoid loading
+// provider modules at startup. Must match packages/ai/src/registry/lm-studio.ts
+// and packages/ai/src/registry/vllm.ts.
 const DEFAULT_LOCAL_TOKEN = "lm-studio-local";
+const DEFAULT_VLLM_LOCAL_TOKEN = "vllm-local";
 
 const SPECIAL_MODEL_MANAGER_PROVIDER_IDS: readonly string[] = [
 	"google-antigravity",
@@ -84,6 +84,10 @@ export function isAuthenticated(apiKey: string | undefined | null): apiKey is st
 	return Boolean(apiKey) && apiKey !== kNoAuth;
 }
 
+function isDiscoveryBearerApiKey(apiKey: string | undefined | null): apiKey is string {
+	return isAuthenticated(apiKey) && apiKey !== DEFAULT_LOCAL_TOKEN && apiKey !== DEFAULT_VLLM_LOCAL_TOKEN;
+}
+
 /** Provider override config (baseUrl, headers, apiKey, compat, transport) without custom models */
 interface ProviderOverride {
 	baseUrl?: string;
@@ -104,9 +108,19 @@ interface ProviderOverride {
  *      `token-plan-sgp.xiaomimimo.com` at discovery time)
  *   3. Existing bundled baseUrl (the host baked into `models.json`)
  *
+ * `transport` resolution priority:
+ *   1. `providerOverride.transport` (e.g. `pi-native` for auth-gateway users)
+ *   2. `existing.transport` (carried over from boot-time override application)
+ *   3. `model.transport` (rarely set — discovery defaults omit it)
+ *
  * Without (1), the user's override would lose to discovery; without (2)
  * preferred over (3), the bundled `api.xiaomimimo.com` would shadow the
  * tp- token-plan host and produce 401s on the first stream call.
+ * Without explicit transport propagation, an openrouter (or any) entry
+ * marked `transport: pi-native` in models.yml silently reverts to the
+ * default openai-completions transport after the background catalog
+ * refresh — so the first `/model` switch after boot hits the raw OpenAI
+ * chat-completions URL instead of the gateway's `/v1/pi/stream` (#2555).
  * See `xiaomi-tp-discovery-merge.test.ts` and the `refresh()` baseUrl-override
  * regression in `model-registry.test.ts`.
  */
@@ -120,6 +134,7 @@ export function mergeDiscoveredModel<TApi extends Api>(
 			...model,
 			baseUrl: providerOverride?.baseUrl ?? model.baseUrl ?? existing.baseUrl,
 			headers: existing.headers ? { ...existing.headers, ...model.headers } : model.headers,
+			transport: providerOverride?.transport ?? existing.transport ?? model.transport,
 			compat: model.compatConfig,
 		} as ModelSpec<TApi>);
 	}
@@ -530,9 +545,8 @@ function finalizeCustomModel(model: CustomModelOverlay, options: CustomModelBuil
 		thinking: resolvedModel.thinking ?? reference?.thinking,
 		input: input as ("text" | "image")[],
 		cost,
-		contextWindow:
-			resolvedModel.contextWindow ?? reference?.contextWindow ?? (options.useDefaults ? 128000 : undefined),
-		maxTokens: resolvedModel.maxTokens ?? reference?.maxTokens ?? (options.useDefaults ? 16384 : undefined),
+		contextWindow: resolvedModel.contextWindow ?? reference?.contextWindow ?? (options.useDefaults ? 128000 : null),
+		maxTokens: resolvedModel.maxTokens ?? reference?.maxTokens ?? (options.useDefaults ? 16384 : null),
 		headers: resolvedModel.headers,
 		omitMaxOutputTokens: resolvedModel.omitMaxOutputTokens ?? reference?.omitMaxOutputTokens,
 		compat: mergeCompat(reference?.compatConfig, resolvedModel.compat),
@@ -866,11 +880,8 @@ export class ModelRegistry {
 			if (!existing) return replacementModel;
 			return {
 				...replacementModel,
-				contextWindow:
-					replacementModel.contextWindow === UNK_CONTEXT_WINDOW
-						? existing.contextWindow
-						: replacementModel.contextWindow,
-				maxTokens: replacementModel.maxTokens === UNK_MAX_TOKENS ? existing.maxTokens : replacementModel.maxTokens,
+				contextWindow: replacementModel.contextWindow ?? existing.contextWindow,
+				maxTokens: replacementModel.maxTokens ?? existing.maxTokens,
 			};
 		});
 	}
@@ -895,6 +906,18 @@ export class ModelRegistry {
 		});
 	}
 
+	#resolveStartupModelCacheProviderId(providerId: string): string {
+		const descriptor = PROVIDER_DESCRIPTORS.find(candidate => candidate.providerId === providerId);
+		if (!descriptor) {
+			return providerId;
+		}
+		const baseUrl =
+			this.#runtimeProviderOverrides.get(providerId)?.baseUrl ??
+			this.#providerOverrides.get(providerId)?.baseUrl ??
+			this.getProviderBaseUrl(providerId);
+		return descriptor.createModelManagerOptions({ baseUrl, fetch: this.#fetch }).cacheProviderId ?? providerId;
+	}
+
 	#loadCachedStandardProviderModels(): { models: Model<Api>[]; authoritativeFreshProviders: Set<string> } {
 		const configuredDiscoveryProviders = new Set(this.#discoverableProviders.map(provider => provider.provider));
 		const cachedModels: Model<Api>[] = [];
@@ -903,7 +926,8 @@ export class ModelRegistry {
 			if (configuredDiscoveryProviders.has(providerId)) {
 				continue;
 			}
-			const cache = readModelCache<Api>(providerId, 24 * 60 * 60 * 1000, Date.now, this.#cacheDbPath);
+			const cacheProviderId = this.#resolveStartupModelCacheProviderId(providerId);
+			const cache = readModelCache<Api>(cacheProviderId, 24 * 60 * 60 * 1000, Date.now, this.#cacheDbPath);
 			if (!cache) {
 				continue;
 			}
@@ -933,7 +957,12 @@ export class ModelRegistry {
 	#loadCachedDiscoverableModels(): Model<Api>[] {
 		const cachedModels: Model<Api>[] = [];
 		for (const providerConfig of this.#discoverableProviders) {
-			const cache = readModelCache<Api>(providerConfig.provider, 24 * 60 * 60 * 1000, Date.now, this.#cacheDbPath);
+			const cache = readModelCache<Api>(
+				this.#configuredDiscoveryCacheProviderId(providerConfig),
+				24 * 60 * 60 * 1000,
+				Date.now,
+				this.#cacheDbPath,
+			);
 			if (!cache) {
 				this.#providerDiscoveryStates.set(providerConfig.provider, {
 					provider: providerConfig.provider,
@@ -1195,11 +1224,19 @@ export class ModelRegistry {
 		this.#rebuildCanonicalIndex();
 	}
 
+	#configuredDiscoveryCacheProviderId(providerConfig: DiscoveryProviderConfig): string {
+		if (providerConfig.discovery.type === "openai-models-list") {
+			return `${providerConfig.provider}:openai-models-list-context-v2`;
+		}
+		return providerConfig.provider;
+	}
+
 	async #discoverProviderModels(
 		providerConfig: DiscoveryProviderConfig,
 		strategy: ModelRefreshStrategy,
 	): Promise<Model<Api>[]> {
-		const cached = readModelCache<Api>(providerConfig.provider, 24 * 60 * 60 * 1000, Date.now, this.#cacheDbPath);
+		const cacheProviderId = this.#configuredDiscoveryCacheProviderId(providerConfig);
+		const cached = readModelCache<Api>(cacheProviderId, 24 * 60 * 60 * 1000, Date.now, this.#cacheDbPath);
 		const requiresAuth = !this.#keylessProviders.has(providerConfig.provider);
 		if (requiresAuth) {
 			const apiKey = await this.#peekApiKeyForProvider(providerConfig.provider);
@@ -1237,6 +1274,7 @@ export class ModelRegistry {
 			providerId,
 			staticModels: [],
 			cacheDbPath: this.#cacheDbPath,
+			cacheProviderId,
 			cacheTtlMs: 24 * 60 * 60 * 1000,
 			fetchDynamicModels,
 		});
@@ -1278,7 +1316,9 @@ export class ModelRegistry {
 			fetch: this.#fetch,
 			getBearerApiKeyResolver: async provider => {
 				const apiKey = await this.getApiKeyForProvider(provider);
-				if (!apiKey || apiKey === DEFAULT_LOCAL_TOKEN || apiKey === kNoAuth) return undefined;
+				if (!isDiscoveryBearerApiKey(apiKey)) {
+					return undefined;
+				}
 				return this.resolver(provider);
 			},
 		};
@@ -1383,11 +1423,20 @@ export class ModelRegistry {
 		for (let i = 0; i < standardProviderDescriptors.length; i++) {
 			const descriptor = standardProviderDescriptors[i];
 			const apiKey = standardProviderKeys[i];
-			if (isAuthenticated(apiKey) || descriptor.allowUnauthenticated) {
+			const hasExplicitVllmConfig =
+				descriptor.providerId === "vllm" &&
+				(this.#runtimeProviderOverrides.has(descriptor.providerId) ||
+					this.#providerOverrides.has(descriptor.providerId) ||
+					this.#keylessProviders.has(descriptor.providerId));
+			if (isAuthenticated(apiKey) || descriptor.allowUnauthenticated || hasExplicitVllmConfig) {
+				const discoveryBaseUrl =
+					this.#runtimeProviderOverrides.get(descriptor.providerId)?.baseUrl ??
+					this.#providerOverrides.get(descriptor.providerId)?.baseUrl ??
+					this.getProviderBaseUrl(descriptor.providerId);
 				options.push(
 					descriptor.createModelManagerOptions({
-						apiKey: isAuthenticated(apiKey) ? apiKey : undefined,
-						baseUrl: this.getProviderBaseUrl(descriptor.providerId),
+						apiKey: isDiscoveryBearerApiKey(apiKey) ? apiKey : undefined,
+						baseUrl: discoveryBaseUrl,
 						fetch: this.#fetch,
 					}),
 				);
@@ -1739,11 +1788,20 @@ export class ModelRegistry {
 	 * paths that pre-flight auth before model resolution) can probe a model
 	 * without resolving an API key. Returns true for keyless providers as well
 	 * as providers with stored credentials. See issue #993.
+	 *
+	 * Side-effect-free and synchronous: a command-backed key (`!cmd`) counts as
+	 * configured by its presence alone — the program is NOT executed — and OAuth
+	 * tokens are NOT refreshed (`authStorage.hasAuth`). This is what keeps the
+	 * model-switch pre-flight off the event loop's hot path; the real key
+	 * (command execution + OAuth refresh) is resolved lazily per request via
+	 * {@link ModelRegistry.resolver}.
 	 */
 	hasConfiguredAuth(model: Model<Api>): boolean {
-		const commandKey = this.#resolveCommandBackedApiKey(model.provider);
+		const keyConfig = this.#customProviderApiKeys.get(model.provider);
 		return (
-			commandKey.configured || this.#keylessProviders.has(model.provider) || this.authStorage.hasAuth(model.provider)
+			isCommandConfigValue(keyConfig) ||
+			this.#keylessProviders.has(model.provider) ||
+			this.authStorage.hasAuth(model.provider)
 		);
 	}
 

package/src/config/model-resolver.ts

@@ -1247,7 +1247,7 @@ export function resolveCliModel(options: {
 			model: undefined,
 			selector: undefined,
 			warning: undefined,
-			error: `Unknown provider "${cliProvider}". Use --list-models to see available providers/models.`,
+			error: `Unknown provider "${cliProvider}". Run "omp models" to see available providers/models.`,
 		};
 	}
 
@@ -1337,7 +1337,7 @@ export function resolveCliModel(options: {
 			selector: undefined,
 			thinkingLevel: undefined,
 			warning,
-			error: `Model "${display}" not found. Use --list-models to see available models.`,
+			error: `Model "${display}" not found. Run "omp models" to see available models.`,
 		};
 	}
 

package/src/config/models-config-schema.ts

@@ -1,4 +1,4 @@
-import * as z from "zod/v4";
+import { z } from "zod/v4";
 
 const OpenRouterRoutingSchema = z.object({
 	only: z.array(z.string()).optional(),
@@ -35,6 +35,7 @@ const OpenAICompatFieldsSchema = z.object({
 	allowsSyntheticReasoningContentForToolCalls: z.boolean().optional(),
 	requiresAssistantContentForToolCalls: z.boolean().optional(),
 	supportsToolChoice: z.boolean().optional(),
+	supportsForcedToolChoice: z.boolean().optional(),
 	disableReasoningOnForcedToolChoice: z.boolean().optional(),
 	disableReasoningOnToolChoice: z.boolean().optional(),
 	thinkingFormat: z.enum(["openai", "openrouter", "zai", "qwen", "qwen-chat-template"]).optional(),
@@ -44,7 +45,7 @@ const OpenAICompatFieldsSchema = z.object({
 	cacheControlFormat: z.enum(["anthropic"]).optional(),
 	supportsStrictMode: z.boolean().optional(),
 	toolStrictMode: z.enum(["all_strict", "none"]).optional(),
-	streamIdleTimeoutMs: z.number().positive().optional(),
+	streamIdleTimeoutMs: z.number().nonnegative().optional(),
 	supportsLongPromptCacheRetention: z.boolean().optional(),
 	supportsReasoningParams: z.boolean().optional(),
 	alwaysSendMaxTokens: z.boolean().optional(),
@@ -124,6 +125,7 @@ const ModelDefinitionSchema = z.object({
 			"azure-openai-responses",
 			"anthropic-messages",
 			"google-generative-ai",
+			"google-gemini-cli",
 			"google-vertex",
 		])
 		.optional(),
@@ -192,6 +194,7 @@ const ProviderConfigSchema = z.object({
 			"azure-openai-responses",
 			"anthropic-messages",
 			"google-generative-ai",
+			"google-gemini-cli",
 			"google-vertex",
 		])
 		.optional(),

package/src/config/models-config.ts

@@ -50,12 +50,13 @@ export function validateProviderConfiguration(
 				!config.headers &&
 				!config.compat &&
 				!config.apiKey &&
+				config.auth !== "none" &&
 				!config.disableStrictTools &&
 				!hasModelOverrides &&
 				!config.discovery
 			) {
 				throw new Error(
-					`Provider ${providerName}: must specify "baseUrl", "headers", "apiKey", "compat", "disableStrictTools", "modelOverrides", "discovery", or "models"`,
+					`Provider ${providerName}: must specify "baseUrl", "headers", "apiKey", "auth: none", "compat", "disableStrictTools", "modelOverrides", "discovery", or "models"`,
 				);
 			}
 		}