@linxiraos/pi-catalog 1.1.27 → 1.2.0

package/CHANGELOG.md

@@ -1,25 +1,12 @@
-# Changelog
-
 ## [Unreleased]
 
-## [1.1.27] - 2026-10-07
-
-- 版本线推进至 1.1.27；本版无独立用户可见变化。
-
-### Added
+## [1.2.0] - 2026-10-12
 
-- Added an optional `statefulResponses` compat field for OpenAI Responses models, kept through OpenRouter's Responses dispatch ([#13686](https://github.com/can1357/oh-my-pi/pull/13686) by [@alphastorm](https://github.com/alphastorm)).
-- Added 15 Snowflake Cortex models with account-specific endpoints, Cortex compatibility rules, and estimated account-billed pricing ([#14507](https://github.com/can1357/oh-my-pi/pull/14507) by [@jorgoose](https://github.com/jorgoose)).
-- Added the `image-tokenization` axis, which declares how GPT-5.2+, Claude and Gemini 3 lines bill input images on every host, with wire-API fallback rules for other models, plus `imageTokens()` to price one image ([#14286](https://github.com/can1357/oh-my-pi/pull/14286) by [@will-bogusz](https://github.com/will-bogusz)).
+- 版本线推进至 1.2.0；本版无独立用户可见变化。
 
-### Changed
-
-- Changed OpenAI Responses endpoints other than OpenAI, Azure OpenAI, and Codex (custom and local servers, proxies including `azure`/`openai-codex` providers pointed at a non-Azure/non-Codex `baseUrl`, OpenRouter) to default `supportsImageDetailOriginal` to `false`, so snapcompact frames and computer screenshots go out as `detail: "auto"` instead of failing on servers that reject `original`; set `compat.supportsImageDetailOriginal: true` to opt a host in ([#13687](https://github.com/can1357/oh-my-pi/pull/13687) by [@alphastorm](https://github.com/alphastorm)).
-- Muse Code can now store Responses results on Meta's side (`store-responses`), so a turn whose connection drops can be recovered instead of re-run. Storage is opt-in via the omp setting `providers.muse-code.storeResponses` or `PI_MUSE_STORE_RESPONSES=1` ([#14293](https://github.com/can1357/oh-my-pi/pull/14293) and [#14534](https://github.com/can1357/oh-my-pi/pull/14534) by [@abilliontokens](https://github.com/abilliontokens)).
-
-## [1.1.25] - 2026-10-03
+## [1.1.27] - 2026-10-07
 
-- 版本线推进至 1.1.25；本版无独立用户可见变化。
+- 版本线推进至 1.1.27；本版无独立用户可见变化。
 
 ## [1.1.24] - 2026-10-03
 

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" | "commandcode" | "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" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "local" | "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" | "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" | "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" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "local" | "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" | "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" | "commandcode" | "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" | "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" | "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" | "exa" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google-antigravity" | "google-gemini-cli" | "helmcode" | "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" | "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/judgment-questions.d.ts

@@ -0,0 +1,7 @@
+import type { Model } from "../types.js";
+/**
+ * Most questions one System One judgment request to this model may carry
+ * (`judgment-questions-max` axis), or `undefined` when the endpoint takes any
+ * batch size.
+ */
+export declare function resolveJudgmentQuestionsMax(model: Model): number | undefined;

package/dist/types/compat/prompt-cache-lookback.d.ts

@@ -0,0 +1,7 @@
+import type { Model } from "../types.js";
+/**
+ * Block positions the provider's prompt cache checks back from a breakpoint for
+ * an earlier request's entry (`prompt-cache-lookback` axis), or `undefined` when
+ * the model has no known lookback bound.
+ */
+export declare function resolvePromptCacheLookback(model: Model): number | undefined;

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

@@ -1,2 +1,2 @@
 /** Every chat-model provider with a catalog entry (`default-model` in its provider rules). */
-export type KnownProvider = "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" | "commandcode" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "helmcode" | "huggingface" | "kilo" | "kimi-code" | "litellm" | "lm-studio" | "local" | "meta" | "minimax" | "minimax-code" | "minimax-code-cn" | "mistral" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai" | "openai-codex" | "opencode-go" | "opencode-zen" | "openrouter" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "snowflake" | "stepfun" | "synthetic" | "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" | "zenmux" | "zhipu-coding-plan";
+export type KnownProvider = "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" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "helmcode" | "huggingface" | "kilo" | "kimi-code" | "litellm" | "lm-studio" | "local" | "meta" | "minimax" | "minimax-code" | "minimax-code-cn" | "mistral" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai" | "openai-codex" | "opencode-go" | "opencode-zen" | "openrouter" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "snowflake" | "stepfun" | "synthetic" | "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" | "zenmux" | "zhipu-coding-plan";

package/dist/types/compat/reasoning-carry.d.ts

@@ -0,0 +1,21 @@
+import type { ModelIdentity } from "./types.js";
+/**
+ * The identity a deployment's reasoning is carried under, or `undefined` when
+ * the deployment's model family is not declared portable there. Class-only
+ * identities never qualify: they name a lineage, not a set of weights.
+ */
+export declare function portableReasoningIdentity(provider: string, api: string, modelId: string): ModelIdentity | undefined;
+/**
+ * Whether reasoning recorded on `source` may replay natively on `target`:
+ * both deployments declare the family portable and classify to the same
+ * class, family, and revision.
+ */
+export declare function carriesReasoning(source: {
+    provider: string;
+    api: string;
+    model: string;
+}, target: {
+    provider: string;
+    api: string;
+    id: string;
+}): boolean;

package/dist/types/compat/request-size.d.ts

@@ -0,0 +1,7 @@
+import type { Model } from "../types.js";
+/**
+ * Maximum encoded inline-image bytes retained per request for a deployment with a documented body-size limit.
+ * The axis is a host contract: a rerouted `baseUrl` (proxy, gateway) imposes its own limit, so the budget
+ * applies only while the model targets its official endpoint.
+ */
+export declare function resolveInlineImageByteBudget(model: Model): number | undefined;

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

@@ -1,13 +1,3 @@
-/**
- * Runtime model-identity taxonomy: classifies wire model ids into
- * `(class, family, revision)` ranks, applies reviewed identity overrides, and
- * exposes the collapse/discovery vocabularies.
- *
- * Faithful port of the o2 reference (`taxonomy.rs` / `classify.rs`):
- * matchers rank `exact(4) > bounded(3) > namespace(2) > prefix(1) > glob(0)`
- * with token byte length as tiebreak; equal cross-class or cross-family ranks
- * throw unless classification is `lenient` (discovery normalization).
- */
 import type { Effort } from "../effort.js";
 import type { ModelIdentity } from "./types.js";
 /** Two classes, families, or reviewed override patterns tie for one model id. */

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

@@ -593,6 +593,8 @@ export interface CompiledSeedModel {
     baseUrl: string;
     reasoning: boolean;
     input: ("text" | "image")[];
+    /** Wire model id sent upstream when it differs from the row id (`request-model-id`). */
+    requestModelId?: string;
     supportsTools?: boolean;
     cost: TokenCost;
     contextWindow: number | null;
@@ -622,6 +624,8 @@ export interface CompiledProvider {
     id: string;
     /** Preferred model id when no explicit selection is made. */
     defaultModel: string;
+    /** Whether the provider participates in automatic default selection (defaults to true). */
+    automaticDefault?: boolean;
     /** Env vars consulted, in order, for the runtime API-key fallback. */
     envVars?: string[];
     /** The runtime creates a model manager even without a valid API key. */

package/dist/types/discovery/antigravity.d.ts

@@ -63,13 +63,23 @@ export interface FetchAntigravityDiscoveryModelsOptions {
      */
     collapseTable?: VariantCollapseTable;
 }
+/**
+ * A complete account roster, or a definitive 401/403 credential rejection.
+ * `null` from {@link fetchAntigravityDiscoveryModels} instead means a transient
+ * failure or invalid response, which must not replace an authoritative catalog.
+ */
+export interface AntigravityModelDiscoveryResult {
+    models: ModelSpec<"google-gemini-cli">[];
+    rejectedStatus?: 401 | 403;
+}
 /**
  * Fetches discoverable Antigravity models and normalizes them into canonical model entries.
  *
- * Returns `null` on network/payload/auth failures.
- * Returns `[]` only when the endpoint responds successfully with no usable models.
+ * Returns `null` on network, server, or payload failures; a 401/403 is reported
+ * separately so multi-account discovery can skip a rejected credential.
+ * A successful empty roster has `models: []` without `rejectedStatus`.
  */
-export declare function fetchAntigravityDiscoveryModels(options: FetchAntigravityDiscoveryModelsOptions): Promise<ModelSpec<"google-gemini-cli">[] | null>;
+export declare function fetchAntigravityDiscoveryModels(options: FetchAntigravityDiscoveryModelsOptions): Promise<AntigravityModelDiscoveryResult | null>;
 /** Advertised image model and serving endpoint for one Antigravity account. */
 export interface AntigravityImageModel {
     id: string;

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

@@ -35,6 +35,22 @@ export interface ModelManagerOptions<TApi extends Api = Api, TModelsDevPayload =
     cacheTtlMs?: number;
     /** When true, a successful dynamic fetch is the complete provider catalog and prunes static-only models. */
     dynamicModelsAuthoritative?: boolean;
+    /**
+     * When true, a provider's dynamic rows are the deployment's whole truth
+     * for input modality: a live text-only row strips the bundled reference's
+     * image support instead of OR-merging it back. Declared by providers whose
+     * capability flags are authoritative and whose endpoints reject
+     * unsupported content (`github-copilot`, `deepinfra`, `coralbricks`).
+     */
+    dynamicInputAuthoritative?: boolean;
+    /** When true, live reasoning capability replaces the fallback, including an explicit false. */
+    dynamicReasoningAuthoritative?: boolean;
+    /**
+     * When true, a live per-million price replaces the fallback even when it is
+     * an explicit `0` (a free model or a dropped cache-write charge). Without it,
+     * a zero discovery price is treated as unknown and the bundled rate is kept.
+     */
+    dynamicCostAuthoritative?: boolean;
     /**
      * When true, a fresh cache never satisfies an online-eligible refresh: the
      * dynamic fetch always runs (an explicit `"offline"` strategy is still

package/dist/types/provider-models/account-access.d.ts

@@ -0,0 +1,11 @@
+import type { Api, ModelSpec } from "../types.js";
+/**
+ * Merge per-account discovery catalogs into one list, deduped by model id.
+ *
+ * The first catalog to list an id supplies its spec; {@link ModelSpec.accountAccess}
+ * is merged from every catalog listing that id, so credential selection can
+ * route account-gated models to the accounts that serve them. Used by
+ * multi-account discovery for Codex and Antigravity, whose rosters differ per
+ * account.
+ */
+export declare function unionAccountCatalogs<TApi extends Api>(catalogs: readonly (readonly ModelSpec<TApi>[])[]): ModelSpec<TApi>[];

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

@@ -11,8 +11,30 @@ export interface GoogleVertexModelManagerConfig {
     signal?: AbortSignal;
     fetch?: FetchImpl;
 }
+/** One Antigravity OAuth account whose `fetchAvailableModels` roster discovery reads. */
+export interface GoogleAntigravityAccount {
+    /** OAuth access token used for `Authorization: Bearer ...`. */
+    accessToken: string;
+    /**
+     * Credential identity recorded in {@link ModelSpec.accountAccess} for every
+     * model this account serves (see `oauthAccountKey` in `@linxiraos/pi-ai`).
+     * When any account lacks one, no model records per-account access.
+     */
+    accountKey?: string;
+}
 export interface GoogleAntigravityModelManagerConfig {
-    oauthToken?: string;
+    /**
+     * Resolves every configured Antigravity account at discovery time. Rosters
+     * are plan-scoped (Claude 5.5 is served to some accounts only) and
+     * authoritative, so each account's roster is fetched and the results are
+     * unioned; reading one account's roster would prune models only its
+     * siblings serve (#14924).
+     *
+     * Returns `null` to abort discovery (e.g. an account's credential failed to
+     * refresh), keeping the previous/bundled catalog instead of caching a
+     * partial one.
+     */
+    resolveAccounts?: () => Promise<readonly GoogleAntigravityAccount[] | null>;
     endpoint?: string;
     fetch?: FetchImpl;
 }

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

@@ -69,7 +69,7 @@ export interface OpenAIModelManagerConfig {
     baseUrl?: string;
     fetch?: FetchImpl;
 }
-export declare function openaiModelManagerOptions(config?: OpenAIModelManagerConfig): ModelManagerOptions<"openai-responses">;
+export declare function openaiModelManagerOptions(config?: OpenAIModelManagerConfig): ModelManagerOptions<Api>;
 /**
  * Re-derive the generated pro-reasoning aliases (`gpt-5.6-*-pro`) for the
  * first-party `openai` gpt-5.6 rows. Each alias inherits the base row's
@@ -154,6 +154,19 @@ export interface DeepinfraModelManagerConfig {
  * model availability at runtime is still gated on credentials by the registry.
  */
 export declare function deepinfraModelManagerOptions(config?: DeepinfraModelManagerConfig): ModelManagerOptions<"openai-completions">;
+export declare const CORALBRICKS_BASE_URL = "https://inference.coralbricks.ai/v1";
+/** CoralBricks OpenAI-compatible discovery configuration. */
+export interface CoralbricksModelManagerConfig {
+    apiKey?: string;
+    baseUrl?: string;
+    fetch?: FetchImpl;
+}
+/**
+ * Builds CoralBricks' model-discovery manager. `/v1/models` is key-protected
+ * (401 without a bearer key), so a keyless config serves only the bundled
+ * reviewed seed rows; with a key, live rows are authoritative over the bundle.
+ */
+export declare function coralbricksModelManagerOptions(config?: CoralbricksModelManagerConfig): ModelManagerOptions<"openai-completions">;
 export interface XaiModelManagerConfig {
     apiKey?: string;
     baseUrl?: string;
@@ -339,6 +352,11 @@ export interface OpenRouterModelManagerConfig {
     baseUrl?: string;
     fetch?: FetchImpl;
 }
+/**
+ * OpenRouter's Decisions API lives at `/api/alpha`, a sibling of the `/api/v1`
+ * chat root; derive it so a custom gateway base URL keeps both aligned.
+ */
+export declare function openrouterDecisionsBaseUrl(chatBaseUrl: string): string;
 export declare function openrouterModelManagerOptions(config?: OpenRouterModelManagerConfig): ModelManagerOptions<Api>;
 export interface ZenMuxModelManagerConfig {
     apiKey?: string;

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

@@ -65,6 +65,8 @@ export declare function devinModelManagerOptions(config?: DevinModelManagerConfi
 export declare function factoryDroidModelManagerOptions(config?: FactoryDroidModelDiscoveryOptions): ModelManagerOptions<"factory-droid-agent">;
 export declare function localModelManagerOptions(): ModelManagerOptions<"local-inference">;
 export declare function webModelManagerOptions(): ModelManagerOptions<"web-search">;
+/** Cloudflare Workers AI judges: the authored Clef rows; the account id resolves per request. */
+export declare function cloudflareWorkersAiModelManagerOptions(): ModelManagerOptions<"workers-ai-systemone">;
 /** Credentials and endpoint overrides for the TypeSafe catalog manager. */
 export interface TypeSafeModelManagerConfig {
     apiKey?: string;

package/dist/types/types.d.ts

@@ -14,7 +14,23 @@ export type KindApiKind = (typeof KIND_API_KINDS)[number];
 /** Grounding transport available to chat models selected by the web role. */
 export type WebSearchGrounding = "gemini" | "anthropic" | "codex" | "xai" | "openrouter" | "openai";
 /** Non-chat runner protocols accepted by catalog seeds, outside the chat dispatch union. */
-export declare const RUNNER_APIS: readonly ["local-inference", "web-search", "typesafe", "openrouter-decisions", "openai-images", "openrouter-images", "xai-tts", "openai-speech", "openai-embeddings", "openrouter-rerank", "openrouter-video", "openai-transcriptions"];
+export declare const RUNNER_APIS: readonly ["local-inference", "web-search", "typesafe", "openrouter-decisions", "openai-decisions", "workers-ai-systemone", "openai-images", "openrouter-images", "xai-tts", "openai-speech", "openai-embeddings", "openrouter-rerank", "openrouter-video", "openai-transcriptions"];
+/** Kind each single-purpose runner API serves; `local-inference` hosts several kinds. */
+export declare const RUNNER_API_KINDS: Record<Exclude<(typeof RUNNER_APIS)[number], "local-inference">, ModelKind>;
+/** Kind a runner API serves; `undefined` for chat transports and multi-kind `local-inference`. */
+export declare function runnerApiKind(api: Api): ModelKind | undefined;
+/** Catalog APIs `generate_image` runs through a pi-ai image client; the hosted Responses pair needs a carrier model. */
+export declare const IMAGE_GENERATION_APIS: readonly ["openai-images", "openrouter-images", "google-generative-ai", "google-gemini-cli", "openai-responses", "openai-codex-responses"];
+export type ImageGenerationApi = (typeof IMAGE_GENERATION_APIS)[number];
+/**
+ * Kinds a model on `api` may declare: a runner api serves its own kind; a chat
+ * transport serves `chat` and `tiny`, plus `image` when `generate_image` runs it
+ * (hosted Responses image tool, Gemini image models). `undefined` for
+ * `local-inference`, which hosts several kinds chosen by the model itself.
+ */
+export declare function servedKinds(api: Api): readonly ModelKind[] | undefined;
+/** Whether a model of `kind` can run on `api`. */
+export declare function apiServesKind(api: Api, kind: ModelKind): boolean;
 /** Resolve a model's kind while preserving chat semantics for existing catalog rows. */
 export declare function modelKind(model: Pick<Model, "kind">): ModelKind;
 /** Canonical thinking transport used by a model. */
@@ -176,6 +192,8 @@ export interface OpenAICompat {
     supportsMultipleSystemMessages?: boolean;
     /** Whether the provider supports `reasoning_effort`. Default: auto-detected from URL. */
     supportsReasoningEffort?: boolean;
+    /** Do not infer a thinking dial when discovery supplies no explicit thinking configuration. */
+    trustExplicitThinkingOnly?: boolean;
     /** Optional mapping from pi-ai reasoning levels to provider/model-specific `reasoning_effort` values. */
     reasoningEffortMap?: Partial<Record<Effort, string>>;
     /** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */
@@ -322,6 +340,8 @@ export interface OpenAICompat {
     extraBody?: Record<string, unknown>;
     /** Request-session header that should mirror the normalized prompt-cache key. Default: unset. */
     promptCacheSessionHeader?: "x-grok-conv-id";
+    /** Request header carrying the conversation session id, independent of prompt caching. Default: unset. */
+    sessionHeader?: "x-litellm-session-id";
     /** Whether chat-completions payloads should include provider-specific prompt-cache markers. */
     cacheControlFormat?: "anthropic" | undefined;
     /**
@@ -405,8 +425,8 @@ export interface OpenAICompat {
     /**
      * Whether the Responses endpoint accepts `configuration_update` input items
      * that change `reasoning.effort` mid-conversation while the request-level
-     * effort stays pinned for prompt caching (GPT-6 Astra). Default:
-     * rule-detected (`true` for `gpt-6-astra` on any host, `false` otherwise).
+     * effort stays pinned for prompt caching (GPT-6 family). Default:
+     * rule-detected (`true` for GPT-6 models on any host, `false` otherwise).
      * Set `false` for custom `openai-responses` / `openai-codex-responses`
      * endpoints that reject the item type with HTTP 400; effort changes are then
      * sent as the top-level `reasoning.effort`.
@@ -603,12 +623,31 @@ export interface AnthropicCompat {
     injectClaudeCodeInstruction?: boolean;
     /** Strip image inputs before encoding (text-only serving of a multimodal id). */
     stripImageInput?: boolean;
+    /**
+     * Largest width or height a single image block may carry. The canonical
+     * Anthropic API rejects anything larger with `At least one of the image
+     * dimensions exceed max allowed size: 8000 pixels`, and the rejection
+     * poisons every retry and model fallback because the block stays in
+     * context, so the request builder downscales to this value first. Hosts
+     * with a different image contract override it. Default: 8000.
+     */
+    maxImageDimension?: number;
+    /**
+     * Largest base64 payload a single image block may carry, in bytes. The
+     * canonical Anthropic API measures the encoded string, not the decoded
+     * bytes (`image exceeds 10 MB maximum: 14012300 bytes > 10485760 bytes`),
+     * and an image inside {@link maxImageDimension} can still cross it, so
+     * payload size is clamped on its own. Hosts with a different image
+     * contract override it. Default: 10485760.
+     */
+    maxImagePayloadBytes?: number;
     /** Thinking-loop watchdog guard family applied to streamed reasoning. */
     thinkingLoopGuard?: "gemini" | "deepseek" | "xai";
     /**
      * Drop the enabled `thinking` config and replayed thinking blocks when the
      * conversation is not thinking-led (a proxy contract for non-interleaved
-     * budget models). Default: false.
+     * budget models), and replay history without thinking blocks on every turn
+     * that sends `thinking: { type: "disabled" }`. Default: false.
      */
     stripThinkingHistory?: boolean;
     /**
@@ -618,10 +657,15 @@ export interface AnthropicCompat {
     effortBeta?: boolean;
     /**
      * Wire form of a disabled-thinking turn: omit the field, send
-     * `{ type: "disabled" }`, or keep adaptive thinking. Unset keeps the
-     * direct-provider behavior.
+     * `{ type: "disabled" }`, keep adaptive thinking, or send
+     * `{ type: "between_tools" }`. Unset keeps the direct-provider behavior.
+     */
+    disabledThinking?: "omit" | "disabled" | "adaptive" | "between-tools";
+    /**
+     * `output_config.effort` pinned on a `between_tools` turn. Unset leaves
+     * the turn's effort untouched.
      */
-    disabledThinking?: "omit" | "disabled" | "adaptive";
+    betweenToolsEffort?: "low" | "medium" | "high";
     /** The model is a fast-mode SKU: send `speed: "fast"` with the fast-mode beta. Default: false. */
     fastMode?: boolean;
 }
@@ -630,6 +674,8 @@ export interface AnthropicCompat {
  * deliberately not used to infer these request-shape capabilities.
  */
 export interface BedrockCompat {
+    /** Explicit disabled-thinking wire form; unset preserves the provider's existing behavior. */
+    disabledThinking?: AnthropicCompat["disabledThinking"];
     /** Whether this endpoint accepts no checkpoints, automatic caching, or explicit cachePoint blocks. */
     promptCacheMode?: "none" | "automatic" | "explicit";
     /** Whether this wire may revise already-streamed text (`stream-revision` axis). Unassigned: append-only. */
@@ -665,9 +711,19 @@ export interface BedrockCompat {
      * choices to `auto` when this is false. Default: true.
      */
     supportsForcedToolChoice?: boolean;
+    /**
+     * Whether the endpoint accepts thinking-binding controls (`thinking.block_binding`,
+     * the `thinking-binding-controls-2026-08-01` beta and the `/input_transformations`
+     * response path) on prefix-bound models. Only consulted when
+     * `thinking.prefixBinding` is set; `false` omits the controls while keeping the
+     * local prefix-binding safeguards. Default: true.
+     */
+    supportsThinkingBindingControls?: boolean;
 }
 /** Fully-resolved Bedrock Converse prompt-cache capabilities, materialized once by `buildModel`. */
 export interface ResolvedBedrockCompat {
+    /** See {@link BedrockCompat.disabledThinking}. */
+    disabledThinking?: BedrockCompat["disabledThinking"];
     promptCacheMode: NonNullable<BedrockCompat["promptCacheMode"]>;
     /** See {@link BedrockCompat.streamRevision}. */
     streamRevision?: BedrockCompat["streamRevision"];
@@ -677,6 +733,8 @@ export interface ResolvedBedrockCompat {
     /** See {@link BedrockCompat.supportsSamplingParams}. */
     supportsSamplingParams?: boolean;
     supportsForcedToolChoice: boolean;
+    /** See {@link BedrockCompat.supportsThinkingBindingControls}. */
+    supportsThinkingBindingControls: boolean;
     /**
      * Stream-watchdog idle-timeout fallback in ms for hosts with no keepalive
      * events; 0 disables the idle watchdog. Undefined defers to
@@ -722,6 +780,7 @@ export interface ResolvedOpenAISharedCompat {
     supportsDeveloperRole: boolean;
     supportsStrictMode: boolean;
     supportsReasoningEffort: boolean;
+    trustExplicitThinkingOnly?: boolean;
     reasoningEffortMap: Partial<Record<Effort, string>>;
     supportsReasoningParams: boolean;
     supportsSamplingParams: boolean;
@@ -764,6 +823,7 @@ export interface ResolvedOpenAISharedCompat {
     emptyLengthFinishIsContextError: boolean;
     usesOpenAIToolCallIdLimit: boolean;
     promptCacheSessionHeader?: OpenAICompat["promptCacheSessionHeader"];
+    sessionHeader?: OpenAICompat["sessionHeader"];
     /**
      * Whether this model accepts explicit OpenAI prompt-cache breakpoints.
      * Built catalog models always materialize this false-by-default value;
@@ -798,7 +858,7 @@ 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" | "statefulResponses" | "supportsSamplingParams" | "supportsPenaltyAndStopParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "disableReasoningWithTools" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "mistralReasoningContentParts" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "syntheticReasoningContentFallback" | "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" | "streamRevision" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "supportsConfigurationUpdate" | "supportsSteering" | "stripImageInput" | "thinkingLoopGuard" | "whenThinking">> & {
+export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "trustExplicitThinkingOnly" | "reasoningEffortMap" | "supportsReasoningParams" | "supportsReasoningSummary" | "statefulResponses" | "supportsSamplingParams" | "supportsPenaltyAndStopParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "disableReasoningWithTools" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "mistralReasoningContentParts" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "syntheticReasoningContentFallback" | "replayReasoningContent" | "qwenPreserveThinking" | "qwenTemplateReasoningEffort" | "requiresThinkingAsText" | "requiresMistralToolIds" | "requiresToolResultName" | "requiresAssistantAfterToolResult" | "requiresAssistantContentForToolCalls" | "stripDeepseekSpecialTokens" | "streamMarkupHealingPattern" | "reasoningDeltasMayBeCumulative" | "emptyLengthFinishIsContextError" | "usesOpenAIToolCallIdLimit" | "promptCacheSessionHeader" | "sessionHeader" | "supportsPromptCacheBreakpoints" | "promptCacheBreakpointTtl" | "openRouterRouting" | "isOpenRouterHost" | "supportsStrictMode" | "supportsLongPromptCacheRetention" | "alwaysSendMaxTokens" | "clampOutputToModelMax" | "wireModelIdMode" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "toolSchemaFlavor" | "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "streamRevision" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "supportsConfigurationUpdate" | "supportsSteering" | "stripImageInput" | "thinkingLoopGuard" | "whenThinking">> & {
     vercelGatewayRouting?: OpenAICompat["vercelGatewayRouting"];
     extraBody?: OpenAICompat["extraBody"];
     cacheControlFormat?: OpenAICompat["cacheControlFormat"];
@@ -828,8 +888,8 @@ export interface ResolvedOpenAIResponsesCompat extends ResolvedOpenAISharedCompa
     /**
      * Whether a `configuration_update` input item may change `reasoning.effort`
      * mid-conversation while the request-level effort stays byte-stable for
-     * prompt caching. Rule-owned: GPT-6 Astra only; every other model rejects
-     * the item type with 400.
+     * prompt caching. Rule-owned: the GPT-6 family; earlier models do not
+     * support the item type.
      */
     supportsConfigurationUpdate: boolean;
     /**
@@ -855,6 +915,13 @@ export interface ResolvedOpenAIResponsesCompat extends ResolvedOpenAISharedCompa
      * `PI_MUSE_STORE_RESPONSES`); stored runs retain prompts and outputs on the provider.
      */
     storeResponses: boolean;
+    /**
+     * Whether the host binds native history items (`encrypted_content`, item
+     * ids) to the connection that issued them, so a new process must rebuild
+     * prior turns from message content until its first successful response.
+     * Rule-owned: GitHub Copilot.
+     */
+    connectionBoundNativeHistory: boolean;
     streamIdleTimeoutMs?: number;
     vercelGatewayRouting?: OpenAICompat["vercelGatewayRouting"];
     /** The model sits behind Vercel AI Gateway's Responses endpoint. */
@@ -878,15 +945,28 @@ export interface ResolvedOpenAIResponsesCompat extends ResolvedOpenAISharedCompa
  */
 export type ResolvedOpenRouterCompat = ResolvedOpenAICompat & ResolvedOpenAIResponsesCompat;
 /** Fully-resolved anthropic-messages compat view (same contract as `ResolvedOpenAICompat`). */
-export type ResolvedAnthropicCompat = Required<Omit<AnthropicCompat, "streamIdleTimeoutMs" | "thinkingLoopGuard" | "bedrockMessagesApi" | "effortBeta" | "disabledThinking" | "stripThinkingHistory" | "fastMode">> & {
+export type ResolvedAnthropicCompat = Required<Omit<AnthropicCompat, "streamIdleTimeoutMs" | "thinkingLoopGuard" | "bedrockMessagesApi" | "effortBeta" | "disabledThinking" | "betweenToolsEffort" | "stripThinkingHistory" | "fastMode" | "maxImageDimension" | "maxImagePayloadBytes">> & {
     /** Effort-beta override; undefined keeps the transport's legacy heuristic. */
     effortBeta?: AnthropicCompat["effortBeta"];
     /** Disabled-thinking wire form; undefined keeps the direct-provider behavior. */
     disabledThinking?: AnthropicCompat["disabledThinking"];
-    /** Strip thinking history on non-thinking-led turns; undefined behaves as false. */
+    /** Effort pinned on `between_tools` turns; undefined leaves effort untouched. */
+    betweenToolsEffort?: AnthropicCompat["betweenToolsEffort"];
+    /** Strip thinking history on non-thinking-led and disabled-thinking turns; undefined behaves as false. */
     stripThinkingHistory?: AnthropicCompat["stripThinkingHistory"];
     /** Fast-mode SKU; undefined behaves as false. */
     fastMode?: AnthropicCompat["fastMode"];
+    /**
+     * Largest width or height a single image block may carry. Undefined defers
+     * to the canonical Anthropic API limit applied by the request builder.
+     */
+    maxImageDimension?: number;
+    /**
+     * Largest base64 image payload a single block may carry, in bytes.
+     * Undefined defers to the canonical Anthropic API limit applied by the
+     * request builder.
+     */
+    maxImagePayloadBytes?: number;
     /** Thinking-loop watchdog guard family applied to streamed reasoning. */
     thinkingLoopGuard?: AnthropicCompat["thinkingLoopGuard"];
     /**
@@ -1090,6 +1170,12 @@ export interface Model<TApi extends Api = Api> {
     id: string;
     /** Role-specific runner capability; omitted for ordinary chat models. */
     kind?: ModelKind;
+    /**
+     * Verbatim configured kind (models.yml, `modelOverrides`, runtime
+     * registrations). `buildModel` applies it over catalog `kind` rules on every
+     * rebuild, so a configured runner keeps its role.
+     */
+    kindConfig?: ModelKind;
     /** Grounding transport supported by this chat model. */
     webSearch?: WebSearchGrounding;
     /** Cheaper same-provider model to run hosted web search in this model's place (model id or provider/id). */
@@ -1187,11 +1273,11 @@ export interface Model<TApi extends Api = Api> {
     cursorMaxModeRoutes?: Readonly<Record<string, boolean>>;
     /**
      * Per-account availability recorded by multi-account discovery: provider
-     * account id (Codex: ChatGPT `chatgpt_account_id`) → that account's
-     * entitlements on this model. An account appears only when its own catalog
-     * lists the model, so credential selection can route account-gated models
-     * (e.g. `gpt-daybreak-blue-latest`) straight to eligible accounts. Absent on
-     * bundled/config rows and on single-account discovery.
+     * account key (Codex: ChatGPT `chatgpt_account_id`; Antigravity: login
+     * email) → that account's entitlements on this model. An account appears
+     * only when its own catalog lists the model, so credential selection can
+     * route account-gated models (e.g. `gpt-daybreak-blue-latest`, Antigravity
+     * Claude 5.5) straight to eligible accounts. Absent on bundled/config rows.
      */
     accountAccess?: Readonly<Record<string, ModelAccountAccess>>;
     /** Cursor `RequestedModel.parameters` for this model's default variant. */
@@ -1302,9 +1388,9 @@ export interface Model<TApi extends Api = Api> {
     /**
      * Per-service-tier cost multipliers baked from the `service-tier-cost`
      * catalog axis (e.g. `{ priority: 2.5 }`). Absent tiers use the API-generic
-     * defaults.
+     * defaults, and a tier with no published price stays at 1x.
      */
-    serviceTierCost?: Readonly<Partial<Record<"flex" | "priority", number>>>;
+    serviceTierCost?: Readonly<Partial<Record<"flex" | "priority" | "ultrafast", number>>>;
     /**
      * Provider-supplied one-line blurb for this model. Set only when an upstream
      * ships one (Devin's `GetCliModelConfigs`); never synthesized locally.
@@ -1314,7 +1400,7 @@ export interface Model<TApi extends Api = Api> {
     isNew?: boolean;
     /** Upstream marks this model as beta / preview quality. */
     isBeta?: boolean;
-    /** Authenticated provider catalog marks this as the account's default model. */
+    /** Authenticated catalog marks this as the account's default; the startup pick prefers it over `default-model`. */
     isProviderDefault?: boolean;
     /** Upstream marks this model as one of its recommended picks. */
     isRecommended?: boolean;

package/dist/types/wire/cloudflare-ai-gateway.d.ts

@@ -6,13 +6,3 @@ export declare const CLOUDFLARE_AI_GATEWAY_ANTHROPIC_BASE_URL = "https://gateway
 export declare const CLOUDFLARE_AI_GATEWAY_OPENAI_BASE_URL = "https://gateway.ai.cloudflare.com/v1/<account>/<gateway>/openai";
 /** OpenAI-compatible Workers AI endpoint. */
 export declare const CLOUDFLARE_AI_GATEWAY_COMPAT_BASE_URL = "https://gateway.ai.cloudflare.com/v1/<account>/<gateway>/compat";
-/** Opaque API-key credential payload persisted by Cloudflare AI Gateway login. */
-export interface CloudflareAiGatewayCredential {
-    token: string;
-    accountId?: string;
-    gatewayId?: string;
-}
-/** Parse both structured login credentials and legacy plain gateway tokens. */
-export declare function parseCloudflareAiGatewayCredential(value: string): CloudflareAiGatewayCredential | null;
-/** Serialize the gateway token with the routing identifiers collected during login. */
-export declare function serializeCloudflareAiGatewayCredential(token: string, accountId: string, gatewayId: string): string;

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

@@ -0,0 +1,10 @@
+/** Cloudflare credential persisted by the AI Gateway and Workers AI logins: a bearer token plus the routing ids collected at login. */
+export interface CloudflareCredential {
+    token: string;
+    accountId?: string;
+    gatewayId?: string;
+}
+/** Parses a structured login credential or a legacy plain token; null when empty or malformed. */
+export declare function parseCloudflareCredential(value: string): CloudflareCredential | null;
+/** Serializes a login credential, trimming fields and omitting empty optional ids. */
+export declare function serializeCloudflareCredential(credential: CloudflareCredential): string;

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 = "0.230.0";
+export declare const FACTORY_DROID_CLIENT_VERSION = "0.237.0";
 /** 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. */