@oh-my-pi/pi-catalog 18.16.0 → 18.18.0

package/CHANGELOG.md

@@ -2,6 +2,32 @@
 
 ## [Unreleased]
 
+## [18.18.0] - 2026-11-07
+
+### Fixed
+
+- Fixed Mistral discovery listing embedding, OCR, moderation, transcription, and speech models as chat models; only rows with `capabilities.completion_chat` are offered, with context, image input, and reasoning read from the row ([#18244](https://github.com/can1357/oh-my-pi/pull/18244) by [@VeigaPunk](https://github.com/VeigaPunk))
+- A failed refresh no longer replaces a cached catalog whose models carry request headers with an empty snapshot; the next launch refetches it instead ([#18227](https://github.com/can1357/oh-my-pi/issues/18227))
+
+## [18.17.0] - 2026-11-06
+
+### Added
+
+- Added `tool-name-prefix` compat axis for Responses hosts that reject certain tool names; Muse Code sets it to `_` ([#17435](https://github.com/can1357/oh-my-pi/issues/17435), [#17515](https://github.com/can1357/oh-my-pi/pull/17515) by [@harshav167](https://github.com/harshav167), continued in [#18053](https://github.com/can1357/oh-my-pi/pull/18053) by [@abilliontokens](https://github.com/abilliontokens))
+- Added `supports-message-threads` compat axis (`supportsMessageThreads`), enabled for Claude 4.5+ on first-party Anthropic
+- Added opt-in Responses WebSocket capability via OMP settings or SDK preferences, off by default ([#17354](https://github.com/can1357/oh-my-pi/pull/17354) by [@TatsukiMeng](https://github.com/TatsukiMeng))
+- Added Factory Droid Gemini 4 Argon on Global inference for accounts with the feature flag ([#17861](https://github.com/can1357/oh-my-pi/pull/17861) by [@will-bogusz](https://github.com/will-bogusz))
+- Added Factory Droid Muse Spark 1.3 (Global region, `muse_spark_1_3` flag) with low to xhigh effort ([#17862](https://github.com/can1357/oh-my-pi/pull/17862) by [@will-bogusz](https://github.com/will-bogusz))
+
+### Changed
+
+- Updated Factory Droid client version reporting; improved tool-input streaming for Claude via Azure Foundry
+
+### Fixed
+
+- Fixed newly published Azure models missing after catalog refresh ([#18173](https://github.com/can1357/oh-my-pi/issues/18173), [#18174](https://github.com/can1357/oh-my-pi/pull/18174) by [@dunkpi](https://github.com/dunkpi))
+- Fixed Factory Droid hiding models when the live-routing family default yields no usable upstream; explicit empty per-model routes still hide them ([#17862](https://github.com/can1357/oh-my-pi/pull/17862) by [@will-bogusz](https://github.com/will-bogusz))
+
 ## [18.16.0] - 2026-11-02
 
 ### 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" | "apple" | "azure" | "baseten" | "bedrock-mantle" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "cloudflare-workers-ai" | "commandcode" | "coralbricks" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "helmcode" | "huggingface" | "inception" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "local" | "meta" | "minimax" | "minimax-code" | "minimax-code-cn" | "mistral" | "modular" | "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" | "sambanova" | "scaleway" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "snowflake" | "stencil" | "stepfun" | "synthetic" | "tavily" | "together" | "typesafe" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "web" | "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" | "apple" | "azure" | "baseten" | "bedrock-mantle" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "cloudflare-workers-ai" | "commandcode" | "coralbricks" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "devin-cli" | "devin-windsurf" | "exa" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "helmcode" | "huggingface" | "inception" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "local" | "meta" | "minimax" | "minimax-code" | "minimax-code-cn" | "mistral" | "modular" | "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" | "sambanova" | "scaleway" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "snowflake" | "stencil" | "stepfun" | "synthetic" | "tavily" | "together" | "typesafe" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "web" | "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" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "cloudflare-workers-ai" | "commandcode" | "coralbricks" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google-antigravity" | "google-gemini-cli" | "helmcode" | "huggingface" | "inception" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "meta" | "minimax-code" | "minimax-code-cn" | "mistral" | "modular" | "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" | "sambanova" | "scaleway" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "snowflake" | "stencil" | "stepfun" | "synthetic" | "tavily" | "together" | "typesafe" | "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" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "cloudflare-workers-ai" | "commandcode" | "coralbricks" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "devin-cli" | "devin-windsurf" | "exa" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google-antigravity" | "google-gemini-cli" | "helmcode" | "huggingface" | "inception" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "meta" | "minimax-code" | "minimax-code-cn" | "mistral" | "modular" | "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" | "sambanova" | "scaleway" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "snowflake" | "stencil" | "stepfun" | "synthetic" | "tavily" | "together" | "typesafe" | "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/responses-websocket.d.ts

@@ -0,0 +1,6 @@
+import type { Model } from "../types.js";
+/**
+ * Capability is rule-owned; callers must opt in. Either explicit false vetoes
+ * the other preference, including a per-request true for a disabled model.
+ */
+export declare function resolveResponsesWebSocketPreference(model: Model, requestPreference?: boolean): boolean;

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

@@ -729,6 +729,12 @@ export interface RequestPolicy {
     /** Default `tool_choice` to `auto` when the request carries tools and the caller picks none. */
     responsesToolChoiceAuto?: boolean;
     googleThinking?: "level" | "level-medium";
+    /** Send no `temperature`, `topP` or `topK` in `generationConfig`. */
+    googleOmitSampling?: boolean;
+    /** Send `generationConfig.maxOutputTokens` as the model's output maximum. */
+    googlePinMaxOutputTokens?: boolean;
+    /** Replay the model's own turns as the native Gemini content it streamed. */
+    googleNativeReplay?: boolean;
     /** The route locks its upstream for the session (`x-provider-routing-source: session_lock`). */
     routingSessionLock?: boolean;
 }

package/dist/types/discovery/devin-proto.d.ts

@@ -1297,6 +1297,21 @@ export interface DevinPlanInfo extends ProtoMessage {
     accountDisplayName: string;
 }
 export declare const DevinPlanInfoSchema: MessageCodec<DevinPlanInfo>;
+/** Devin message exa.seat_management_pb.ExchangePKCEAuthorizationCodeRequest. */
+export interface ExchangePKCEAuthorizationCodeRequest extends ProtoMessage {
+    authorizationCode: string;
+    codeVerifier: string;
+}
+export declare const ExchangePKCEAuthorizationCodeRequestSchema: MessageCodec<ExchangePKCEAuthorizationCodeRequest>;
+/** Devin message exa.seat_management_pb.ExchangePKCEAuthorizationCodeResponse. */
+export interface ExchangePKCEAuthorizationCodeResponse extends ProtoMessage {
+    apiKey: string;
+    name: string;
+    apiServerUrl: string;
+    devinWebappHost: string;
+    devinApiUrl: string;
+}
+export declare const ExchangePKCEAuthorizationCodeResponseSchema: MessageCodec<ExchangePKCEAuthorizationCodeResponse>;
 /** Devin message exa.codeium_common_pb.ExperimentConfig. */
 export interface ExperimentConfig extends ProtoMessage {
     experiments: ExperimentWithVariant[];

package/dist/types/model-manager.d.ts

@@ -110,6 +110,8 @@ export interface ModelResolutionResult<TApi extends Api = Api> {
     stale: boolean;
     source: ModelResolutionSource;
     updatedAt?: number;
+    /** Why the provider's endpoint fetcher threw this call; the models then come from cache, models.dev or bundled data. */
+    fetchError?: string;
 }
 /**
  * Stateful facade over provider model resolution.

package/dist/types/types.d.ts

@@ -493,6 +493,14 @@ export interface AnthropicCompat {
      * Sonnet 4.6+, Fable/Mythos 5+ on supported hosts. Default: false.
      */
     supportsServerCompaction?: boolean;
+    /**
+     * Whether the model supports message threads (`message-threads-2026-08-12`):
+     * later turns continue a server-held conversation and send only new
+     * messages. Enabled on Claude 4.5+ for the first-party provider; the
+     * transport still requires the official endpoint unless the caller opts in.
+     * Default: false.
+     */
+    supportsMessageThreads?: boolean;
     /**
      * Whether the model is served by the first-party Anthropic provider (its
      * default route is the official API). Rule-owned on the provider; the
@@ -955,6 +963,14 @@ export interface ResolvedOpenAIResponsesCompat extends ResolvedOpenAISharedCompa
      * `PI_MUSE_STORE_RESPONSES`); stored runs retain prompts and outputs on the provider.
      */
     storeResponses: boolean;
+    /**
+     * Prefix the `openai-responses` handler puts on every function-tool name it
+     * declares or replays, and strips from returned calls through the request's
+     * wire-name table. Rule-owned: hosts that reject requests declaring certain
+     * exact tool names (Muse Code). Unset sends names unchanged. Other
+     * Responses-family handlers do not read it.
+     */
+    toolNamePrefix?: string;
     /**
      * Whether the host binds native history items (`encrypted_content`, item
      * ids) to the connection that issued them, so a new process must rebuild
@@ -1466,7 +1482,7 @@ export interface Model<TApi extends Api = Api> {
      * dispatch is redirected.
      */
     transport?: "pi-native";
-    /** Hint that websocket transport should be preferred when supported by the provider implementation. */
+    /** Standard Responses is opt-in and catalog-gated; either model/request false vetoes true. Codex keeps its independent policy. */
     preferWebsockets?: boolean;
     /** Codex Responses Lite transport: send the lite marker and carry instructions/tools as input items (mirrors codex-rs `use_responses_lite`). */
     useResponsesLite?: boolean;

package/dist/types/wire/codex.d.ts

@@ -46,7 +46,6 @@ export declare const OPENAI_HEADERS: {
     readonly ROUTING_HINT: "x-codex-routing-hint";
 };
 export declare const OPENAI_HEADER_VALUES: {
-    readonly BETA_RESPONSES: "responses=experimental";
     readonly BETA_RESPONSES_WEBSOCKETS_V2: "responses_websockets=2026-02-06";
     readonly REMOTE_COMPACTION_V2: "remote_compaction_v2";
     readonly ORIGINATOR_CODEX: "omp";

package/dist/types/wire/factory-droid.d.ts

@@ -3,7 +3,7 @@
  * native request corpus with `packages/ai/scripts/capture-factory-droid-native.ts`
  * against the matching CLI binary and make `factory-droid-native-parity.test.ts` pass.
  */
-export declare const FACTORY_DROID_CLIENT_VERSION = "0.237.0";
+export declare const FACTORY_DROID_CLIENT_VERSION = "1.1.0";
 /** Wire protocols the Factory proxy multiplexes; `api-routes provider="factory-droid"` picks one per model. */
 export declare const FACTORY_DROID_WIRES: readonly ["openai-completions", "openai-responses", "anthropic-messages", "google-generate"];
 /** Wire protocol used by the Factory proxy for each model. */

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "18.16.0",
+	"version": "18.18.0",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"keywords": [
 		"ai",
@@ -78,7 +78,7 @@
 	},
 	"scripts": {
 		"check": "oxlint . && oxfmt --check --no-error-on-unmatched-pattern 'src/**/*.{ts,tsx}' '{test,bench,examples,scripts}/**/*.ts' '*.ts' && bun run check:types",
-		"check:types": "tsgo -p tsconfig.json --noEmit",
+		"check:types": "bun check --no-script -p tsconfig.json",
 		"lint": "oxlint .",
 		"test": "bun test --parallel",
 		"fix": "oxlint --fix --fix-suggestions . && bun run fmt",
@@ -88,12 +88,12 @@
 		"gen:proto": "bun scripts/generate-protocols.ts"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.16.0",
-		"@oh-my-pi/pi-utils": "18.16.0"
+		"@oh-my-pi/omptype": "18.18.0",
+		"@oh-my-pi/pi-utils": "18.18.0"
 	},
 	"devDependencies": {
 		"@bgotink/kdl": "0.4.0",
-		"@oh-my-pi/pi-ai": "18.16.0",
+		"@oh-my-pi/pi-ai": "18.18.0",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/auth-ids.ts

@@ -25,6 +25,8 @@ export type AuthProviderId =
 	| "deepinfra"
 	| "deepseek"
 	| "devin"
+	| "devin-cli"
+	| "devin-windsurf"
 	| "exa"
 	| "factory-droid"
 	| "firepass"
@@ -123,6 +125,8 @@ export type LoginProviderId =
 	| "deepinfra"
 	| "deepseek"
 	| "devin"
+	| "devin-cli"
+	| "devin-windsurf"
 	| "exa"
 	| "factory-droid"
 	| "firepass"

package/src/compat/axes.ts

@@ -126,6 +126,9 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"responses-tool-choice-auto": wire("responsesToolChoiceAuto", ["request"]),
 	"routing-session-lock": wire("routingSessionLock", ["request"]),
 	"google-thinking": wire("googleThinking", ["request"], "scalar", ["level", "level-medium"]),
+	"google-omit-sampling": wire("googleOmitSampling", ["request"]),
+	"google-pin-max-output-tokens": wire("googlePinMaxOutputTokens", ["request"]),
+	"google-native-replay": wire("googleNativeReplay", ["request"]),
 	// ── wire: OpenAI-compatible surfaces (chat completions + Responses) ──
 	"allows-synthetic-reasoning-content-for-tool-calls": wire("allowsSyntheticReasoningContentForToolCalls", OAI),
 	"always-send-max-tokens": wire("alwaysSendMaxTokens", OAI),
@@ -207,6 +210,7 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"supports-reasoning-params": wire("supportsReasoningParams", OAI),
 	"supports-reasoning-summary": wire("supportsReasoningSummary", ["openai-responses"]),
 	"store-responses": wire("storeResponses", ["openai-responses"]),
+	"tool-name-prefix": wire("toolNamePrefix", ["openai-responses"], "scalar", ["_"]),
 	"supports-store": wire("supportsStore", ["openai"]),
 	"supports-strict-mode": wire("supportsStrictMode", OAI),
 	"supports-tool-choice": wire("supportsToolChoice", OAI),
@@ -271,6 +275,7 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"supports-mid-conversation-tool-changes": wire("supportsMidConversationToolChanges", ["anthropic"]),
 	"supports-per-message-effort": wire("supportsPerMessageEffort", ["anthropic"]),
 	"supports-server-compaction": wire("supportsServerCompaction", ["anthropic", "bedrock"]),
+	"supports-message-threads": wire("supportsMessageThreads", ["anthropic"]),
 	"supports-between-tools-thinking": wire("supportsBetweenToolsThinking", ["anthropic"]),
 	"supports-thinking-binding-controls": wire("supportsThinkingBindingControls", ["anthropic", "bedrock"]),
 	"supports-turn-scoped-system": wire("supportsTurnScopedSystem", ["anthropic"]),
@@ -339,6 +344,7 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"thinking-unverified": { key: "unverified", set: "thinking", shape: "scalar" },
 
 	// ── catalog metadata ──
+	"supports-websockets": { key: "supportsWebsockets", set: "catalog", shape: "scalar", values: [true, false] },
 	"apply-patch-tool-type": {
 		key: "applyPatchToolType",
 		set: "catalog",

package/src/compat/resolve.ts

@@ -870,6 +870,7 @@ function pickResponsesOnly(compat: ResolvedOpenAIResponsesCompat): ResponsesOnly
 		supportsImageDetailOriginal: compat.supportsImageDetailOriginal,
 		supportsObfuscationOptOut: compat.supportsObfuscationOptOut,
 		storeResponses: compat.storeResponses,
+		toolNamePrefix: compat.toolNamePrefix,
 		connectionBoundNativeHistory: compat.connectionBoundNativeHistory,
 		supportsAllTurnsReasoningContext: compat.supportsAllTurnsReasoningContext,
 		supportsConfigurationUpdate: compat.supportsConfigurationUpdate,
@@ -903,6 +904,7 @@ function resolveAnthropicPolicy(
 		signingEndpoint,
 		supportsContextManagement: true,
 		supportsServerCompaction: false,
+		supportsMessageThreads: false,
 		firstPartyProvider: false,
 		honorsClaudeCodeBetaKillSwitch: false,
 		supportsOutputEffort: true,

package/src/compat/responses-websocket.ts

@@ -0,0 +1,12 @@
+import type { Model } from "../types";
+import { resolveCatalogPolicy } from "./catalog-policy";
+
+/**
+ * Capability is rule-owned; callers must opt in. Either explicit false vetoes
+ * the other preference, including a per-request true for a disabled model.
+ */
+export function resolveResponsesWebSocketPreference(model: Model, requestPreference?: boolean): boolean {
+	if (resolveCatalogPolicy(model).supportsWebsockets !== true) return false;
+	if (requestPreference === false || model.preferWebsockets === false) return false;
+	return requestPreference === true || model.preferWebsockets === true;
+}

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

@@ -2,7 +2,7 @@
 // (and does not set `show-in-login-list #false`) must appear here; providers
 // without a login flow sort alphabetically after this list.
 login-order "openai-codex" "anthropic" "zai" "zai-coding-plan" "kimi-code" "openrouter" "github-copilot" \
-	"cursor" "devin" "google-antigravity" "google-gemini-cli" "openai-codex-device" "xai" "xai-oauth" \
+	"cursor" "devin" "devin-cli" "devin-windsurf" "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" "sambanova" "minimax-code" "minimax-code-cn" "xiaomi" "xiaomi-token-plan-sgp" \
 	"xiaomi-token-plan-ams" "xiaomi-token-plan-cn" "firepass" "cline-pass" "factory-droid" "commandcode" "charm-hyper" "deepseek" "stepfun" "muse-code" "meta" "moonshot" "inception" \

package/src/compat/rules/auth/devin-cli.kdl

@@ -0,0 +1,6 @@
+auth "devin-cli" {
+	name "Devin (adopt CLI credentials)"
+	login "custom" hook="devin-cli"
+	refresh "none"
+	store-as "devin"
+}

package/src/compat/rules/auth/devin-windsurf.kdl

@@ -0,0 +1,6 @@
+auth "devin-windsurf" {
+	name "Devin (Windsurf Enterprise)"
+	login "custom" hook="devin-windsurf"
+	refresh "none"
+	store-as "devin"
+}

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

@@ -209,6 +209,22 @@ class "anthropic" {
 			stops-output-at-context-window #true
 		}
 	}
+	// Message threads (`message-threads-2026-08-12`) keep the conversation on
+	// the server so later turns send only new messages. Claude 4.5 and newer
+	// support them on the Claude API; the transport checks the effective URL
+	// per request because `ANTHROPIC_BASE_URL` / Foundry reroutes leave this
+	// flag stale. Vertex and Bedrock do not support them.
+	// https://platform.claude.com/docs/en/build-with-claude/message-threads
+	on "anthropic" {
+		on-api "anthropic-messages" {
+			revision ">=4.5" {
+				supports-message-threads #true
+			}
+			models "claude-mythos-preview" {
+				supports-message-threads #true
+			}
+		}
+	}
 	// Opus 5.5 rejects forced tool selection the way Fable/Mythos do — a
 	// `tool_choice` of `tool` or `any` fails the request with 400
 	// `invalid_request_error`: `tool_choice: type "tool" and "any" are not

package/src/compat/rules/providers/factory-droid.kdl

@@ -279,6 +279,12 @@ provider "factory-droid" {
 			cost input=0 output=0 cache-read=0 cache-write=0
 			limits context=1000000 max-tokens=65536
 		}
+		model "gemini-4-argon" name="Gemini 4 Argon" {
+			reasoning #true
+			input "text" "image"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits context=934464 max-tokens=65536
+		}
 		model "gemini-3.1-pro-preview" name="Gemini 3.1 Pro" {
 			reasoning #true
 			input "text" "image"
@@ -393,6 +399,12 @@ provider "factory-droid" {
 			cost input=0 output=0 cache-read=0 cache-write=0
 			limits context=908928 max-tokens=131072
 		}
+		model "muse-spark-1.3" name="Muse Spark 1.3" {
+			reasoning #true
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits context=908928 max-tokens=131072
+		}
 		model "beam-1.0" name="Beam 1.0" {
 			reasoning #true
 			input "text"
@@ -937,6 +949,18 @@ provider "factory-droid" {
 		thinking-default-level "high"
 		entitlement { feature-flag "garnet_0715" }
 	}
+	// Launches Global-only: Factory serves it from neither US nor EU inference.
+	models "gemini-4-argon" {
+		routing-family "google"
+		upstream-rotation "google"
+		region-upstreams-us
+		credit-rates { input 1.6; output 5; cache-read 0.05 }
+		list-price-from "google"
+		thinking-efforts "low" "medium" "high"
+		thinking-requires-effort #true
+		thinking-default-level "medium"
+		entitlement { feature-flag "gemini_4_argon" }
+	}
 	models "gemini-3.1-pro-preview" {
 		routing-family "google"
 		upstream-rotation "google"
@@ -1171,6 +1195,19 @@ provider "factory-droid" {
 		thinking-requires-effort #true
 		thinking-default-level "high"
 	}
+	// Vertex Model Garden serves 1.3 on its global endpoint only and rejects the
+	// `none` and `max` tiers; the ladder outranks the Meta class ladder as 1.2's does.
+	models "muse-spark-1.3" priority=1 {
+		routing-family "factory"
+		upstream-rotation "google"
+		region-upstreams-us
+		credit-rates { input 0.5; output 3.4; cache-read 0.12 }
+		list-price-from "meta"
+		thinking-efforts "low" "medium" "high" "xhigh"
+		thinking-requires-effort #true
+		thinking-default-level "high"
+		entitlement { feature-flag "muse_spark_1_3" }
+	}
 	models "beam-1.0" {
 		routing-family "factory"
 		upstream-rotation "baseten"
@@ -1448,9 +1485,9 @@ provider "factory-droid" {
 			server-side-fallback-models "claude-opus-5"
 		}
 	}
-	// With tools, Anthropic and Vertex stream tool input eagerly and advertise
-	// fine-grained tool streaming; Bedrock takes the beta alone.
-	on-upstream "anthropic" "vertex_anthropic" {
+	// With tools, Anthropic, Vertex and Azure Foundry stream tool input eagerly
+	// and advertise fine-grained tool streaming; Bedrock takes the beta alone.
+	on-upstream "anthropic" "vertex_anthropic" "azure_anthropic" {
 		supports-eager-tool-input-streaming #true
 		anthropic-tool-streaming-beta #true
 	}
@@ -1515,12 +1552,27 @@ provider "factory-droid" {
 	models "grok-*" {
 		supports-reasoning-summary #false
 	}
+	// Muse Spark 1.3 rides Responses through Vertex. Native asks for encrypted
+	// reasoning but never replays it: Factory-served models fall outside the
+	// reasoning it carries across Responses turns. It requests the model's full
+	// output cap, above the generic Responses clamp.
+	models "muse-spark-1.3" {
+		filter-reasoning-history #true
+		clamp-output-to-model-max #true
+	}
 	models "gemini-*" "garnet-07-15" {
 		omit-max-output-tokens #true
 		google-thinking "level"
 	}
-	models "gemini-3.1-pro-preview" "gemini-3.7-flash" "gemini-3.8-flash" {
+	models "gemini-4-argon" "gemini-3.1-pro-preview" "gemini-3.7-flash" "gemini-3.8-flash" {
 		google-thinking "level-medium"
 	}
+	// Argon's request policy: no sampling parameters, the output cap pinned
+	// to the model maximum, and its own model turns replayed as Gemini sent them.
+	models "gemini-4-argon" {
+		google-omit-sampling #true
+		google-pin-max-output-tokens #true
+		google-native-replay #true
+	}
 	// FACTORY_API_KEY is a control-plane key, not a subscription proxy token.
 }

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

@@ -67,7 +67,23 @@ provider "muse-code" {
 	// `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"
+	// A Muse Code subscription key answers 403 `client_attestation_denied` when
+	// the request declares a tool with certain exact, case-sensitive names:
+	// `bash`, `shell`, `web_search`, `web_fetch`, `apply_patch`, `read_file`
+	// and `update_plan` (measured 2026-10-31, still 2026-11-03; `Bash`,
+	// `run_shell`, `local_shell` and omp's other built-in names pass, and the
+	// list is not exhaustive). Only declared names count: a replayed `bash`
+	// call in history is accepted. Prefixing every function tool with `_` was
+	// accepted with all of omp's built-in tools declared. Meta pay-as-you-go
+	// keys are not attested and keep plain names.
+	tool-name-prefix "_"
 	on-api "openai-responses" {
+		// api.meta.ai answers 400 "name must be at most 64 characters" for a longer
+		// tool name (#9130), and omp caps MCP names at exactly 64 for it. Scoped by
+		// provider and API so it outranks the 128 Responses default in
+		// tool-call-names.kdl. A name the `_` prefix would push past it goes out
+		// unprefixed.
+		tool-call-name-max-length 64
 		// Spark 1.3 Contributor warmed siblings hit 10/10 vs 0/10 with per-child keys (2026-10-12).
 		// Meta documents shared-prefix prompt_cache_key routing; keep the opt-in on these exact Muse Code IDs.
 		models "muse-spark-1.3" "muse-spark-1.3-contributor" "muse-wonder-1.0" "muse-wonder-1.0-contributor" "muse-wonder-full-1.0" "muse-wonder-full-1.0-contributor" {

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

@@ -4,6 +4,9 @@ provider "openai" {
 	default-model "gpt-5.5"
 	env "OPENAI_API_KEY"
 	image-count-budget 200
+	on-api "openai-responses" {
+		supports-websockets #true
+	}
 	kind-apis {
 		embedding "openai-embeddings"
 		image "openai-responses"

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

@@ -48,6 +48,13 @@ provider "web" {
 			limits
 			supports-tools #false
 		}
+		model "keenable" name="Keenable" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
 		model "jina" name="Jina" {
 			reasoning #false
 			input "text"

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

@@ -109,10 +109,10 @@ behavior {
 			"claude-sonnet-4-6" "claude-sonnet-4-5-20250929" "claude-haiku-5-5" "claude-haiku-4-5-20251001" "gpt-6-astra" \
 			"gpt-6.1-sol" "gpt-6-sol" "gpt-6-luna" "gpt-5.6-sol" "gpt-5.6-sol-fast" "gpt-5.6-terra" \
 			"gpt-6-astra-ultrafast" "gpt-6.1-sol-ultrafast" "gpt-6.1-terra" "gpt-6-sol-fast" "gpt-6-luna-fast" \
-			"muse-spark-open" "muse-spark-1.2" \
+			"muse-spark-open" "muse-spark-1.2" "muse-spark-1.3" \
 			"gpt-5.6-luna" "gpt-5.5" "gpt-5.5-fast" "gpt-5.5-pro" "gpt-5.4" "gpt-5.4-fast" "gpt-5.4-mini" \
 			"gpt-5.4-mini-fast" "gpt-5.3-codex" "gpt-5.3-codex-fast" "gpt-5.2" "garnet-07-15" \
-			"gemini-3.1-pro-preview" "gemini-3.8-flash" "gemini-3.7-flash" "gemini-3.6-flash" \
+			"gemini-4-argon" "gemini-3.1-pro-preview" "gemini-3.8-flash" "gemini-3.7-flash" "gemini-3.6-flash" \
 			"gemini-3.5-flash" "gemini-3-flash-preview" "grok-4.7" "grok-4.6" "grok-4.5" "atlas-07-21" \
 			"aster-07-15" "gpt-image-2.5-sunburst" "gpt-image-2.5-flare" "gemini-3-pro-image-preview"
 	}
@@ -316,13 +316,13 @@ behavior {
 			exact="minimax-m2.7"
 		route "openai-responses" exact="gpt-6-astra" exact="gpt-6.1-sol" exact="gpt-6-sol" exact="gpt-6-luna" \
 			exact="gpt-6-astra-ultrafast" exact="gpt-6.1-sol-ultrafast" exact="gpt-6.1-terra" exact="gpt-6-sol-fast" \
-			exact="gpt-6-luna-fast" \
+			exact="gpt-6-luna-fast" exact="muse-spark-1.3" \
 			exact="gpt-5.6-sol" exact="gpt-5.6-sol-fast" exact="gpt-5.6-terra" exact="gpt-5.6-luna" \
 			exact="gpt-5.5" exact="gpt-5.5-fast" exact="gpt-5.5-pro" exact="gpt-5.4" exact="gpt-5.4-fast" \
 			exact="gpt-5.4-mini" exact="gpt-5.4-mini-fast" exact="gpt-5.3-codex" exact="gpt-5.3-codex-fast" \
 			exact="gpt-5.2" exact="grok-4.7" exact="grok-4.6" exact="grok-4.5" exact="gpt-image-2.5-sunburst" \
 			exact="gpt-image-2.5-flare"
-		route "google-generate" exact="garnet-07-15" exact="gemini-3.1-pro-preview" exact="gemini-3.8-flash" \
+		route "google-generate" exact="garnet-07-15" exact="gemini-4-argon" exact="gemini-3.1-pro-preview" exact="gemini-3.8-flash" \
 			exact="gemini-3.7-flash" exact="gemini-3.6-flash" exact="gemini-3.5-flash" \
 			exact="gemini-3-flash-preview" exact="gemini-3-pro-image-preview"
 		route "openai-completions" exact="inkling" exact="mistral-large-4" exact="mistral-medium-3.5" exact="glm-5.3-flash" \