@oh-my-pi/pi-catalog 18.1.11 → 18.1.13

package/src/provider-models/openai-compat.ts

@@ -1,4 +1,4 @@
-import { USER_AGENT } from "@oh-my-pi/pi-utils";
+import { USER_AGENT, getInstallId } from "@oh-my-pi/pi-utils";
 import * as logger from "@oh-my-pi/pi-utils/logger";
 import { toClinePassPublicModelId } from "../cline-pass-model-id";
 import {
@@ -3056,6 +3056,11 @@ function openCodeModelManagerOptions(
 					provider: providerId,
 					baseUrl: discoveryBaseUrl,
 					apiKey,
+					// Live discovery hits the OpenCode gateway outside any
+					// conversation: attribute with the stable install id
+					// (x-opencode-session required from 09/06) and omp's UA
+					// instead of Bun's default.
+					headers: { "User-Agent": USER_AGENT, "x-opencode-session": getInstallId() },
 					mapModel: (entry, defaults) => {
 						const reference = references.get(defaults.id);
 						const name = toModelName(entry.name, reference?.name ?? defaults.name);
@@ -4360,7 +4365,7 @@ export function coreWeaveModelManagerOptions(
 // 15.75 Meta Model API
 // ---------------------------------------------------------------------------
 
-const META_MODEL_API_BASE_URL = "https://api.meta.ai/v1";
+const META_MODEL_API_BASE_URL = getDefaultModelDiscoveryBaseUrl("meta")!;
 const META_MUSE_SPARK_COST = { input: 1.25, output: 4.25, cacheRead: 0.15, cacheWrite: 0 } as const;
 // Contributor SKUs (`-contributor`): same model, discounted because prompts
 // are used for training.
@@ -4369,6 +4374,12 @@ const META_MUSE_SPARK_THINKING: ThinkingConfig = {
 	mode: "effort",
 	efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh],
 };
+// Meta documents the `max` effort tier for Muse Spark 1.3 (standard) only;
+// contributor tiers and other revisions stay on the 5-tier ladder.
+const META_MUSE_SPARK_MAX_THINKING: ThinkingConfig = {
+	mode: "effort",
+	efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh, Effort.Max],
+};
 
 function museSparkSpec(revision: string, tier: "standard" | "contributor"): ModelSpec<"openai-responses"> {
 	const contributor = tier === "contributor";
@@ -4383,7 +4394,7 @@ function museSparkSpec(revision: string, tier: "standard" | "contributor"): Mode
 		cost: contributor ? META_MUSE_SPARK_CONTRIBUTOR_COST : META_MUSE_SPARK_COST,
 		contextWindow: 1_048_576,
 		maxTokens: 131_072,
-		thinking: META_MUSE_SPARK_THINKING,
+		thinking: revision === "1.3" && tier === "standard" ? META_MUSE_SPARK_MAX_THINKING : META_MUSE_SPARK_THINKING,
 		compat: {
 			supportsReasoningEffort: true,
 			includeEncryptedReasoning: true,
@@ -4572,6 +4583,47 @@ export function metaModelManagerOptions(config?: MetaModelManagerConfig): ModelM
 	};
 }
 
+/** Muse Code shares Meta Model API's model capabilities and equivalent token pricing. */
+export const MUSE_CODE_STATIC_MODELS: readonly ModelSpec<"openai-responses">[] = META_MUSE_STATIC_MODELS.map(model => ({
+	...model,
+	provider: "muse-code",
+}));
+
+const MUSE_CODE_MODEL_BY_ID: Partial<Record<string, ModelSpec<"openai-responses">>> = Object.fromEntries(
+	MUSE_CODE_STATIC_MODELS.map(model => [model.id, model]),
+);
+
+function museCodeLineageSpec(id: string): ModelSpec<"openai-responses"> | undefined {
+	const model = museSparkLineageSpec(id);
+	return model ? { ...model, provider: "muse-code" } : undefined;
+}
+
+export function museCodeModelManagerOptions(config?: MetaModelManagerConfig): ModelManagerOptions<"openai-responses"> {
+	return {
+		...createOpenAICompatibleModelManagerOptions({
+			api: "openai-responses",
+			providerId: "muse-code",
+			defaultBaseUrl: META_MODEL_API_BASE_URL,
+			config,
+			headers: { "x-api-version": "1.0.0" },
+			dynamicModelsAuthoritative: true,
+			requireApiKey: true,
+			filterModel: (_entry, model) => !isExcludedModel("muse-code", model.id),
+			mapModel: (entry, defaults, reference) =>
+				mapWithBundledReference(
+					entry,
+					defaults,
+					reference ?? MUSE_CODE_MODEL_BY_ID[defaults.id] ?? museCodeLineageSpec(defaults.id),
+				),
+		}),
+		cacheProviderId: resolveModelCacheProviderId("muse-code", {
+			apiKey: config?.apiKey,
+			baseUrl: config?.baseUrl ?? META_MODEL_API_BASE_URL,
+		}),
+		staticModels: MUSE_CODE_STATIC_MODELS,
+	};
+}
+
 // ---------------------------------------------------------------------------
 // 16. Moonshot
 // ---------------------------------------------------------------------------
@@ -6021,6 +6073,8 @@ export interface GithubCopilotModelManagerConfig {
 }
 
 const COPILOT_CACHE_INVALIDATED_MODEL_IDS = [
+	"gpt-6-astra",
+	"gpt-6-astra-1m",
 	"grok-4.5",
 	"grok-4.5-1m",
 	"grok-4.6",

package/src/types.ts

@@ -709,6 +709,8 @@ export interface ResolvedOpenAISharedCompat {
 	isOpenRouterHost: boolean;
 	/** Whether this endpoint needs a max-token field even when caller did not set one. */
 	alwaysSendMaxTokens: boolean;
+	/** Clamp a requested output-token count to the model's advertised ceiling. */
+	clampOutputToModelMax: boolean;
 	openRouterRouting?: OpenAICompat["openRouterRouting"];
 	/** Provider-specific wire model-id transform applied to the base id. */
 	wireModelIdMode: "raw" | "cline-pass" | "firepass" | "fireworks" | "openrouter";
@@ -777,6 +779,7 @@ export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat &
 			| "supportsStrictMode"
 			| "supportsLongPromptCacheRetention"
 			| "alwaysSendMaxTokens"
+			| "clampOutputToModelMax"
 			| "wireModelIdMode"
 			| "vercelGatewayRouting"
 			| "extraBody"
@@ -1118,6 +1121,8 @@ export interface Model<TApi extends Api = Api> {
 	/** Premium Copilot requests charged per user-initiated request (defaults to 1). */
 	premiumMultiplier?: number;
 	contextWindow: number | null;
+	/** Optional larger prompt window available when extended context is enabled. */
+	maxContextWindow?: number;
 	maxTokens: number | null;
 	/**
 	 * When `true`, providers MUST omit `max_output_tokens` (Responses) /
@@ -1199,6 +1204,13 @@ export interface Model<TApi extends Api = Api> {
 	 * - `"function"` or undefined: JSON function-tool with `{input: string}` (spec §1.2).
 	 */
 	applyPatchToolType?: "freeform" | "function";
+	/**
+	 * Edit-tool description density for this model. `"compact"` selects the
+	 * terse mode prompt (all operations and invariants preserved) for hosts
+	 * where per-request prompt bytes are the dominant cost. Generated catalog
+	 * policy sets it; the edit tool falls back to the full prompt when unset.
+	 */
+	editPromptVariant?: "full" | "compact";
 	/**
 	 * Force OAuth-style request shaping for providers whose API key prefix doesn't
 	 * match an OAuth token (e.g. routing Anthropic traffic through a proxy that