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

package/CHANGELOG.md

@@ -2,6 +2,22 @@
 
 ## [Unreleased]
 
+## [18.1.12] - 2026-09-06
+
+### Added
+
+- Added Muse Code as a provider with Muse Spark models and live account-scoped discovery.
+- Muse Code subscriptions now resolve a compact edit-prompt variant, cutting recurring per-request tool bytes without touching other providers.
+- Added Meta's new `max` reasoning effort tier to Muse Spark 1.3 (standard) on the Meta Model API and Muse Code.
+
+### Fixed
+
+- Fixed OpenCode Go/Zen live model discovery (`GET /v1/models`) missing `x-opencode-session` and omp's `User-Agent`: discovery requests now attribute with the stable install id so the requests OpenCode flags as `Bun fetch` carry the required session header.
+	- Fixed GPT-6 Astra requests through GitHub Copilot failing with an unsupported endpoint error ([#10874](https://github.com/can1357/oh-my-pi/pull/10874) by [@xpcmdshell](https://github.com/xpcmdshell)).
+	- Fixed GPT-6 Astra showing as free with a 272K-token window in the OpenAI Codex catalog by applying its documented pricing; `/extended-context` enables the wire-advertised 872K-token maximum ([#10980](https://github.com/can1357/oh-my-pi/pull/10980) by [@H4vC](https://github.com/H4vC)).
+	- Fixed GPT-6 Astra compacting early at a 272K-token window with its full window gated behind `/extended-context`: it now defaults to the documented 1.05M-token window.
+	- Made extended-context catalog rebuilds faster by resolving each model's maximum window once per process ([#11039](https://github.com/can1357/oh-my-pi/pull/11039) by [@H4vC](https://github.com/H4vC)).
+
 ## [18.1.9] - 2026-09-04
 
 ### Added

package/dist/types/compat/auth-ids.d.ts

@@ -1,4 +1,4 @@
 /** Every provider with an `auth/<id>.kdl` policy. */
-export type AuthProviderId = "abliteration" | "aiand" | "aimlapi" | "alibaba-coding-plan" | "alibaba-token-plan" | "amazon-bedrock" | "anthropic" | "azure" | "baseten" | "bedrock-mantle" | "cerebras" | "cline-pass" | "cloudflare-ai-gateway" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "huggingface" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "meta" | "minimax" | "minimax-code" | "minimax-code-cn" | "mistral" | "moonshot" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai" | "openai-codex" | "openai-codex-device" | "opencode-go" | "opencode-zen" | "openrouter" | "parallel" | "perplexity" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "synthetic" | "tavily" | "together" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zai-coding-plan" | "zenmux" | "zhipu-coding-plan";
+export type AuthProviderId = "abliteration" | "aiand" | "aimlapi" | "alibaba-coding-plan" | "alibaba-token-plan" | "amazon-bedrock" | "anthropic" | "azure" | "baseten" | "bedrock-mantle" | "cerebras" | "cline-pass" | "cloudflare-ai-gateway" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "huggingface" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "meta" | "minimax" | "minimax-code" | "minimax-code-cn" | "mistral" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai" | "openai-codex" | "openai-codex-device" | "opencode-go" | "opencode-zen" | "openrouter" | "parallel" | "perplexity" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "synthetic" | "tavily" | "together" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zai-coding-plan" | "zenmux" | "zhipu-coding-plan";
 /** Providers whose policy declares a `login` flow (the `/login` roster). */
-export type LoginProviderId = "abliteration" | "aiand" | "alibaba-coding-plan" | "alibaba-token-plan" | "anthropic" | "baseten" | "cerebras" | "cline-pass" | "cloudflare-ai-gateway" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google-antigravity" | "google-gemini-cli" | "huggingface" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "meta" | "minimax-code" | "minimax-code-cn" | "moonshot" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai-codex" | "openai-codex-device" | "opencode-go" | "opencode-zen" | "openrouter" | "parallel" | "perplexity" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "synthetic" | "tavily" | "together" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zai-coding-plan" | "zenmux" | "zhipu-coding-plan";
+export type LoginProviderId = "abliteration" | "aiand" | "alibaba-coding-plan" | "alibaba-token-plan" | "anthropic" | "baseten" | "cerebras" | "cline-pass" | "cloudflare-ai-gateway" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google-antigravity" | "google-gemini-cli" | "huggingface" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "meta" | "minimax-code" | "minimax-code-cn" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai-codex" | "openai-codex-device" | "opencode-go" | "opencode-zen" | "openrouter" | "parallel" | "perplexity" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "synthetic" | "tavily" | "together" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zai-coding-plan" | "zenmux" | "zhipu-coding-plan";

package/dist/types/compat/context-window.d.ts

@@ -0,0 +1,7 @@
+import type { Model } from "../types.js";
+/**
+ * Maximum prompt window for extended context. Live discovery takes precedence
+ * over rule-owned fallbacks for older bundled or cached model metadata. Resolve
+ * at catalog composition time so frozen bundled rows need no runtime mutation.
+ */
+export declare function resolveMaxContextWindow(model: Model): number | undefined;

package/dist/types/compat/types.d.ts

@@ -367,6 +367,7 @@ export type CompiledCredentialExpiry = {
     path: string;
     fromPath?: string;
     skewMs: number;
+    fallbackMs?: number;
 } | {
     mode: "jwt";
     skewMs: number;

package/dist/types/provider-models/descriptors.d.ts

@@ -262,6 +262,11 @@ export declare const CATALOG_PROVIDERS: readonly [{
     readonly defaultModel: "devstral-medium-latest";
     readonly envVars: readonly ["MISTRAL_API_KEY"];
     readonly createModelManagerOptions: (config: ModelManagerConfig) => import("../index.js").ModelManagerOptions<"openai-completions", unknown>;
+}, {
+    readonly id: "muse-code";
+    readonly defaultModel: "muse-spark-1.3";
+    readonly createModelManagerOptions: (config: ModelManagerConfig) => import("../index.js").ModelManagerOptions<"openai-responses", unknown>;
+    readonly dynamicModelsAuthoritative: true;
 }, {
     readonly id: "meta";
     readonly defaultModel: "muse-spark-1.1";

package/dist/types/provider-models/openai-compat.d.ts

@@ -509,6 +509,9 @@ export interface MetaModelManagerConfig {
     fetch?: FetchImpl;
 }
 export declare function metaModelManagerOptions(config?: MetaModelManagerConfig): ModelManagerOptions<"openai-responses">;
+/** Muse Code shares Meta Model API's model capabilities and equivalent token pricing. */
+export declare const MUSE_CODE_STATIC_MODELS: readonly ModelSpec<"openai-responses">[];
+export declare function museCodeModelManagerOptions(config?: MetaModelManagerConfig): ModelManagerOptions<"openai-responses">;
 export interface MoonshotModelManagerConfig {
     apiKey?: string;
     baseUrl?: string;

package/dist/types/types.d.ts

@@ -653,6 +653,8 @@ 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";
@@ -673,7 +675,7 @@ export interface ResolvedOpenAISharedCompat {
  * `buildModel`; request handlers read fields and never detect, resolve, or
  * allocate.
  */
-export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "reasoningEffortMap" | "supportsReasoningParams" | "supportsReasoningSummary" | "supportsSamplingParams" | "supportsPenaltyAndStopParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "replayReasoningContent" | "qwenPreserveThinking" | "qwenTemplateReasoningEffort" | "requiresThinkingAsText" | "requiresMistralToolIds" | "requiresToolResultName" | "requiresAssistantAfterToolResult" | "requiresAssistantContentForToolCalls" | "stripDeepseekSpecialTokens" | "streamMarkupHealingPattern" | "reasoningDeltasMayBeCumulative" | "emptyLengthFinishIsContextError" | "usesOpenAIToolCallIdLimit" | "promptCacheSessionHeader" | "supportsPromptCacheBreakpoints" | "promptCacheBreakpointTtl" | "openRouterRouting" | "isOpenRouterHost" | "supportsStrictMode" | "supportsLongPromptCacheRetention" | "alwaysSendMaxTokens" | "wireModelIdMode" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "toolSchemaFlavor" | "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "stripImageInput" | "thinkingLoopGuard" | "whenThinking">> & {
+export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "reasoningEffortMap" | "supportsReasoningParams" | "supportsReasoningSummary" | "supportsSamplingParams" | "supportsPenaltyAndStopParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "replayReasoningContent" | "qwenPreserveThinking" | "qwenTemplateReasoningEffort" | "requiresThinkingAsText" | "requiresMistralToolIds" | "requiresToolResultName" | "requiresAssistantAfterToolResult" | "requiresAssistantContentForToolCalls" | "stripDeepseekSpecialTokens" | "streamMarkupHealingPattern" | "reasoningDeltasMayBeCumulative" | "emptyLengthFinishIsContextError" | "usesOpenAIToolCallIdLimit" | "promptCacheSessionHeader" | "supportsPromptCacheBreakpoints" | "promptCacheBreakpointTtl" | "openRouterRouting" | "isOpenRouterHost" | "supportsStrictMode" | "supportsLongPromptCacheRetention" | "alwaysSendMaxTokens" | "clampOutputToModelMax" | "wireModelIdMode" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "toolSchemaFlavor" | "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "stripImageInput" | "thinkingLoopGuard" | "whenThinking">> & {
     vercelGatewayRouting?: OpenAICompat["vercelGatewayRouting"];
     extraBody?: OpenAICompat["extraBody"];
     cacheControlFormat?: OpenAICompat["cacheControlFormat"];
@@ -942,6 +944,8 @@ 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) /
@@ -1023,6 +1027,13 @@ 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

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "18.1.11",
+	"version": "18.1.13",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"homepage": "https://omp.sh",
 	"author": "Stencil Labs, Inc.",
@@ -35,12 +35,12 @@
 		"gen:proto": "bun scripts/generate-protocols.ts"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.1.11",
-		"@oh-my-pi/pi-utils": "18.1.11"
+		"@oh-my-pi/omptype": "18.1.13",
+		"@oh-my-pi/pi-utils": "18.1.13"
 	},
 	"devDependencies": {
 		"@bgotink/kdl": "0.4.0",
-		"@oh-my-pi/pi-ai": "18.1.11",
+		"@oh-my-pi/pi-ai": "18.1.13",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/build.ts

@@ -52,6 +52,10 @@ function applyCatalogAssignments<TApi extends Api>(model: Model<TApi>, catalog:
 	if (applyPatchToolType === "freeform" || applyPatchToolType === "function") {
 		model.applyPatchToolType = applyPatchToolType;
 	}
+	const editPromptVariant = catalog.editPromptVariant;
+	if (editPromptVariant === "full" || editPromptVariant === "compact") {
+		model.editPromptVariant = editPromptVariant;
+	}
 	const requiresCursorToolSchemaProjection = catalog.requiresCursorToolSchemaProjection;
 	if (requiresCursorToolSchemaProjection === true) {
 		model.requiresCursorToolSchemaProjection = true;

package/src/compat/auth-ids.ts

@@ -45,6 +45,7 @@ export type AuthProviderId =
 	| "minimax-code-cn"
 	| "mistral"
 	| "moonshot"
+	| "muse-code"
 	| "nanogpt"
 	| "novita"
 	| "nvidia"
@@ -119,6 +120,7 @@ export type LoginProviderId =
 	| "minimax-code"
 	| "minimax-code-cn"
 	| "moonshot"
+	| "muse-code"
 	| "nanogpt"
 	| "novita"
 	| "nvidia"

package/src/compat/axes.ts

@@ -90,7 +90,7 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"allows-synthetic-reasoning-content-for-tool-calls": wire("allowsSyntheticReasoningContentForToolCalls", OAI),
 	"always-send-max-tokens": wire("alwaysSendMaxTokens", OAI),
 	"cache-control-format": wire("cacheControlFormat", OAI, "scalar", ["anthropic"]),
-	"clamp-output-to-model-max": wire("clampOutputToModelMax", ["openai"]),
+	"clamp-output-to-model-max": wire("clampOutputToModelMax", OAI),
 	"disable-reasoning-on-forced-tool-choice": wire("disableReasoningOnForcedToolChoice", OAI),
 	"disable-reasoning-on-tool-choice": wire("disableReasoningOnToolChoice", OAI),
 	"drop-thinking-when-reasoning-effort": wire("dropThinkingWhenReasoningEffort", ["openai"]),
@@ -271,11 +271,13 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"context-window-floor": { key: "contextWindowFloor", set: "catalog", shape: "scalar" },
 	"cost-patch": { key: "costPatch", set: "catalog", shape: "object" },
 	"delegation-bias": { key: "delegationBias", set: "catalog", shape: "scalar", values: DELEGATION_BIASES },
+	"edit-prompt-variant": { key: "editPromptVariant", set: "catalog", shape: "scalar", values: ["full", "compact"] },
 	"edit-revision": { key: "editRevision", set: "catalog", shape: "scalar" },
 	"input-modalities": { key: "inputModalities", set: "catalog", shape: "array", values: ["text", "image"] },
 	"limits-patch": { key: "limitsPatch", set: "catalog", shape: "object" },
 	"long-context-cost": { key: "longContext", set: "catalog", shape: "object" },
 	"long-usage-limit-fallback": { key: "longUsageLimitFallback", set: "catalog", shape: "scalar" },
+	"max-context-window": { key: "maxContextWindow", set: "catalog", shape: "scalar" },
 	"requires-cursor-tool-schema-projection": {
 		key: "requiresCursorToolSchemaProjection",
 		set: "catalog",

package/src/compat/context-window.ts

@@ -0,0 +1,31 @@
+import { toModelSpec } from "../provider-models/bundled-references";
+import type { Model } from "../types";
+import { resolveModelPolicy } from "./resolve";
+
+/**
+ * Rule-owned fallback maxima by `provider/id/api`. The rule table is static
+ * per process (compiled into rules.json), so one resolution per model key is
+ * enough — composition calls this per model on every rebuild while extended
+ * context is enabled, and the full policy resolve (identity classification
+ * plus cascade walk) is wasted work per call.
+ */
+const ruleFallbackCache = new Map<string, number | undefined>();
+
+/**
+ * Maximum prompt window for extended context. Live discovery takes precedence
+ * over rule-owned fallbacks for older bundled or cached model metadata. Resolve
+ * at catalog composition time so frozen bundled rows need no runtime mutation.
+ */
+export function resolveMaxContextWindow(model: Model): number | undefined {
+	const maximum = model.maxContextWindow;
+	if (typeof maximum === "number" && Number.isFinite(maximum) && maximum > 0) {
+		return maximum;
+	}
+
+	const key = `${model.provider} ${model.id} ${model.api}`;
+	if (ruleFallbackCache.has(key)) return ruleFallbackCache.get(key);
+	const fallback = resolveModelPolicy(toModelSpec(model)).catalog.maxContextWindow;
+	const resolved = typeof fallback === "number" && Number.isFinite(fallback) && fallback > 0 ? fallback : undefined;
+	ruleFallbackCache.set(key, resolved);
+	return resolved;
+}

package/src/compat/resolve.ts

@@ -753,6 +753,7 @@ function resolveOpenAIResponsesPolicy(
 		wireModelIdMode: isOpenRouter ? "openrouter" : "raw",
 		toolSchemaFlavor: facts.is("kimi") ? "moonshot-mfjs" : undefined,
 		alwaysSendMaxTokens: facts.is("kimi"),
+		clampOutputToModelMax: false,
 		supportsObfuscationOptOut: isOpenAIUrl || provider === "openai",
 		officialEndpoint: isOfficialOpenAIEndpoint(provider, baseUrl),
 		harmonyLeakMitigation: false,

package/src/compat/rules/README.md

@@ -276,7 +276,7 @@ auth "anthropic" {
         credential {
             access "access_token"                // dot path; `claim="a|b"` reads JWT claims; `literal="…"` pins a value
             refresh "refresh_token"
-            expires "seconds" path="expires_in" from="created_at" skew-ms=300000   // or `expires "jwt" fallback-ms=N` / `expires "never"`
+            expires "seconds" path="expires_in" from="created_at" skew-ms=300000 fallback-ms=3600000   // or `expires "jwt" fallback-ms=N` / `expires "never"`
             email "account.email_address"        // also account-id, org-id, org-name, project-id, api-endpoint, enterprise-url
         }
         userinfo url="https://…/userinfo" email="email" account-id="sub"

package/src/compat/rules/auth/_order.kdl

@@ -5,7 +5,7 @@ login-order "openai-codex" "anthropic" "zai" "zai-coding-plan" "kimi-code" "open
 	"cursor" "devin" "google-antigravity" "google-gemini-cli" "openai-codex-device" "xai" "xai-oauth" \
 	"gitlab-duo" "gitlab-duo-agent" "alibaba-coding-plan" "alibaba-token-plan" "aiand" "abliteration" "zhipu-coding-plan" \
 	"umans" "qwen-portal" "sakana" "minimax-code" "minimax-code-cn" "xiaomi" "xiaomi-token-plan-sgp" \
-	"xiaomi-token-plan-ams" "xiaomi-token-plan-cn" "firepass" "cline-pass" "deepseek" "meta" "moonshot" \
+	"xiaomi-token-plan-ams" "xiaomi-token-plan-cn" "firepass" "cline-pass" "deepseek" "muse-code" "meta" "moonshot" \
 	"cerebras" "baseten" "fireworks" "together" "nvidia" "novita" "deepinfra" "huggingface" "perplexity" \
 	"qianfan" "venice" "siliconflow" "siliconflow-cn" "synthetic" "nanogpt" "wafer-serverless" "coreweave" \
 	"vercel-ai-gateway" "cloudflare-ai-gateway" "litellm" "kilo" "zenmux" "opencode-zen" "opencode-go" \

package/src/compat/rules/auth/muse-code.kdl

@@ -0,0 +1,28 @@
+auth "muse-code" {
+	name "Muse Code (Subscription)"
+	expiry "jwt-or-never"
+	login "device-code" {
+		client-id "1031625952748946"
+		device url="https://auth.meta.com/oidc/device/authorization/" {
+			headers {
+				Accept "application/json"
+				"x-api-version" "1.0.0"
+			}
+		}
+		token url="https://auth.meta.com/oidc/device/token/" {
+			headers {
+				Accept "application/json"
+				"x-api-version" "1.0.0"
+			}
+		}
+		response user-code="user_code" device-code="device_code" verification-uri="verification_uri" verification-uri-complete="verification_uri_complete" interval="interval" expires-in="expires_in"
+		instructions "Enter code: {user_code}"
+		credential {
+			// Model requests use the minted API key, not the account token. Meta's
+			// device response omits expiry and rejects refresh_token grants.
+			expires "never"
+		}
+		after-exchange hook="muse-code-key"
+	}
+	refresh "none"
+}

package/src/compat/rules/classes/meta.kdl

@@ -3,7 +3,13 @@
 class "meta" {
 	// Exact ids on purpose: only an exact selector upgrades hosts whose
 	// discovery misreports a Muse Spark revision as non-reasoning.
-	models "muse-spark-1.2" "muse-spark-1.2-contributor" "muse-spark-1.3" "muse-spark-1.3-contributor" {
+	// muse-spark-1.3 exposes the max effort tier; older/contributor revisions
+	// keep the 5-tier ladder.
+	models "muse-spark-1.3" {
+		thinking-efforts "minimal" "low" "medium" "high" "xhigh" "max"
+		thinking-mode "effort"
+	}
+	models "muse-spark-1.1" "muse-spark-1.2" "muse-spark-1.2-contributor" "muse-spark-1.3-contributor" {
 		thinking-efforts "minimal" "low" "medium" "high" "xhigh"
 		thinking-mode "effort"
 	}

package/src/compat/rules/providers/meta.kdl

@@ -3,6 +3,7 @@
 provider "meta" {
 	include-encrypted-reasoning #true
 	supports-reasoning-effort #true
+	clamp-output-to-model-max #true
 	class "unknown" {
 		thinking-efforts "minimal" "low" "medium" "high" "xhigh"
 		thinking-mode "effort"

package/src/compat/rules/providers/muse-code.kdl

@@ -0,0 +1,16 @@
+// Muse Code subscriptions use the same Model API wire contract as direct Meta API keys.
+
+provider "muse-code" {
+	include-encrypted-reasoning #true
+	supports-reasoning-effort #true
+	clamp-output-to-model-max #true
+	// Subscription tier bills per request; the compact edit prompt cuts the
+	// recurring tool-schema bytes. Verified 2026-09-05 against api.meta.ai/v1:
+	// `custom` tools are rejected with 400 "`custom` tools are not supported
+	// on this endpoint", so apply-patch must stay a function tool here.
+	edit-prompt-variant "compact"
+	class "unknown" {
+		thinking-efforts "minimal" "low" "medium" "high" "xhigh"
+		thinking-mode "effort"
+	}
+}

package/src/compat/rules/providers/openai-codex.kdl

@@ -29,6 +29,29 @@ provider "openai-codex" {
 			cache-write 15.625
 		}
 	}
+	// Codex discovery still reports Astra's stale 272K window with
+	// max_context_window=872000 (client 0.153.1, 2026-09-04) — the same stale
+	// pair it reports for Luna/Sol/Terra. Public docs list a 1.05M window and
+	// live Codex traffic succeeds past the wire maximum (895,302 input tokens
+	// observed, including cached input), so floor to the documented window like
+	// the GPT-5.6 1M floor below. Astra carries no long-context surcharge, so
+	// the window must not be gated behind `/extended-context`; the
+	// `maximum > contextWindow` guard then ignores the stale 872K live maximum.
+	// Pricing stays at the documented credit-equivalent rates, with free cache
+	// writes and no >272K multiplier.
+	models "gpt-6-astra" "gpt-6-astra-wm" {
+		cost-patch {
+			input 10.0
+			output 50.0
+			cache-read 1.0
+			cache-write 0
+		}
+		service-tier-cost {
+			flex 0.5
+			priority 2.5
+		}
+		context-window-floor 1050000
+	}
 	class "openai" {
 		revision ">=5.3 <5.7" {
 			thinking-mode "effort"
@@ -118,4 +141,8 @@ provider "openai-codex" {
 	models "gpt-5.6-luna" "gpt-5.6-sol" "gpt-5.6-terra" {
 		context-window-floor 1000000
 	}
+	// (No `max-context-window` fallback: the 1.05M floor already exceeds the
+	// stale 872K wire maximum, so offline and cached rows compose to the
+	// documented window and the live maximum is ignored by the
+	// `maximum > contextWindow` guard — same as Luna/Sol/Terra above.)
 }

package/src/compat/rules/runtime/behavior.kdl

@@ -127,6 +127,7 @@ behavior {
 	// Meta's /v1/models lists image generation and transcription SKUs beside
 	// the Muse Spark chat models.
 	exclude-models provider="meta" prefix="muse-image-" prefix="muse-voice-"
+	exclude-models provider="muse-code" prefix="muse-image-" prefix="muse-voice-"
 	// NanoGPT's roster interleaves non-text SKUs by name token.
 	exclude-models provider="nanogpt" substring="embedding" substring="image" substring="vision" \
 		substring="audio" substring="speech" substring="transcribe" substring="moderation" \
@@ -154,7 +155,7 @@ behavior {
 	api-routes provider="github-copilot" default="openai-completions" {
 		route "anthropic-messages" glob="claude-haiku-*" glob="claude-sonnet-*" glob="claude-opus-*" \
 			glob="claude-fable-*" glob="claude-mythos-*"
-		route "openai-responses" exact="grok-4.5" exact="grok-4.6" prefix="gpt-5" prefix="oswe" prefix="mai-"
+		route "openai-responses" exact="grok-4.5" exact="grok-4.6" prefix="gpt-5" prefix="gpt-6" prefix="oswe" prefix="mai-"
 	}
 	// ZenMux fronts Anthropic ids on a dedicated Messages endpoint; every
 	// other namespace stays on chat completions.