@kenkaiiii/gg-ai 5.93.0 → 5.95.0

package/dist/index.d.cts

@@ -253,26 +253,18 @@ interface StreamOptions {
     /** OpenAI service tier for latency-sensitive requests. Only sent to first-party OpenAI API calls. */
     serviceTier?: "auto" | "default" | "flex" | "priority";
     /** Provider speed tier. "fast" sends Anthropic `speed: "fast"` (first-party API
-     *  only) or OpenAI/Codex `service_tier: "priority"`. Set automatically by
+     *  only) or OpenAI `service_tier: "priority"`. Set automatically by
      *  `stream()` for fast-mode model ids (see `resolveFastModeModel`). */
     speed?: "standard" | "fast";
-    /** Codex endpoint only: send the Responses-Lite request shape (lite header,
-     *  `parallel_tool_calls: false`, all-turns reasoning context). Unset follows
-     *  the model family (on for gpt-5.6/gpt-6 models). Lite allows only one tool
-     *  call per response, so turning it off lets the model batch calls. */
-    responsesLite?: boolean;
-    /** OpenAI only (Codex and API-key routes): send tools with strict
+    /** OpenAI only (ChatGPT plan and API-key routes): send tools with strict
      *  (grammar-constrained) schemas. Default true. Strict calls carry every
      *  optional field as null, which costs output tokens and latency per call. */
     strictTools?: boolean;
-    /** OpenAI ChatGPT account ID (from OAuth JWT) for codex endpoint */
+    /** ChatGPT plan sign-in: the account subject. Set, it routes OpenAI to the plan. */
     accountId?: string;
-    /** Stable conversation identity for Codex transport headers. This is distinct from
-     *  promptCacheKey: sessions with matching prefixes may share a cache key, but must
-     *  retain independent session/thread identities. */
+    /** Stable conversation identity, distinct from promptCacheKey: sessions with
+     *  matching prefixes may share a cache key but keep independent identities. */
     transportSessionId?: string;
-    /** Google Cloud/Code Assist project ID used by Gemini OAuth transport. */
-    projectId?: string;
     /** Enable provider-native web search. Each provider uses its own format:
      *  - Anthropic: server tool `web_search_20250305`
      *  - Moonshot: `builtin_function` `$web_search`
@@ -315,16 +307,8 @@ interface StreamOptions {
      *  budget_tokens < max_tokens), no request is sent and the response has
      *  zero usage with stopReason "end_turn". */
     prewarm?: boolean;
-    /** Override the User-Agent sent with OAuth-authenticated Anthropic requests.
-     *  Anthropic's OAuth edge rejects requests whose claude-cli version lags too
-     *  far behind the real Claude Code release; callers that track the live
-     *  version should pass it here. Ignored for non-Anthropic providers and for
-     *  Anthropic requests using a regular API key. */
-    userAgent?: string;
-    /** Extra HTTP headers attached to every model request. Used by providers
-     *  whose endpoint gates on client identity (e.g. Kimi For Coding requires a
-     *  `User-Agent: kimi-code-cli/...` and `X-Msh-*` device headers). Merged
-     *  into the underlying SDK's default headers. */
+    /** Extra HTTP headers attached to every model request. Merged into the
+     *  underlying SDK's default headers. */
     defaultHeaders?: Record<string, string>;
     /** Provider-specific request-body fields, merged into the JSON body of every
      *  model request (e.g. OpenRouter's routing object:
@@ -985,8 +969,8 @@ declare const TOOL_CALL_NAME_RULES: Record<ToolCallNameRule["id"], ToolCallNameR
 declare function isValidToolCallName(name: unknown, rule: ToolCallNameRule): boolean;
 /**
  * Pick the tool-name rule for the transport a request will actually use. Mirrors
- * the routing in `stream.ts`: `openai` with an `accountId` is the Codex
- * (Responses) endpoint, MiniMax rides the Anthropic transport but is a third
+ * the routing in `stream.ts`: `openai` itself (the ChatGPT plan or an API key)
+ * is the Responses API, MiniMax rides the Anthropic transport but is a third
  * party, and an `openai` pointed at a custom baseUrl is an unknown compatible
  * server.
  */
@@ -1047,9 +1031,15 @@ declare function continuesTurn(message: Message): boolean;
 /** A real turn boundary: a user message that isn't a runtime notice. */
 declare function opensTurn(message: Message): boolean;
 
-declare const CODEX_CLIENT_VERSION = "0.166.0";
+/**
+ * GPT-6 family (and gpt-5.6): reasoning can't be turned off, and the models
+ * default to low verbosity. Point releases keep the dotted version (gpt-6.1-sol).
+ */
 declare function usesResponsesLite(model: string): boolean;
 
+/** GG Coder's own client identity. Sent where a provider needs a User-Agent; never another vendor's. */
+declare const GG_CODER_USER_AGENT = "ggcoder (+https://github.com/KenKaiii/gg-coder)";
+
 /**
  * Fire a minimal `max_tokens: 1` request that populates the Anthropic prompt
  * cache with the system prompt + tools prefix, so the first real user turn is
@@ -1067,24 +1057,10 @@ declare function prewarmAnthropicCache(options: {
     tools?: StreamOptions["tools"];
     serverTools?: StreamOptions["serverTools"];
     baseUrl?: string;
-    userAgent?: string;
     cacheRetention?: StreamOptions["cacheRetention"];
     signal?: AbortSignal;
 }): Promise<void>;
 
-declare const CLAUDE_CODE_FALLBACK_VERSION = "2.1.316";
-/** Test hook: the pending background refresh, if any. */
-declare function pendingClaudeCodeVersionRefresh(): Promise<void> | null;
-/**
- * Resolve the current Claude Code release version for spoofing the claude-cli
- * User-Agent on OAuth and inference requests. Cached in-memory for the process
- * lifetime and on disk for 24h. Falls back to a hardcoded constant if the npm
- * registry is unreachable and no cache exists.
- */
-declare function getClaudeCodeVersion(): Promise<string>;
-/** Build the User-Agent string Anthropic's OAuth + inference edges expect. */
-declare function getClaudeCliUserAgent(): Promise<string>;
-
 interface PalsuProviderState {
     /** `stream()` calls served. */
     callCount: number;
@@ -1171,4 +1147,4 @@ declare function palsuEmbedding(text: string, dimensions?: number): number[];
 /** Deterministic fake relevance: the share of distinct query words found in the document (0–1). */
 declare function palsuRerankScore(query: string, document: string): number;
 
-export { type AssistantMessage, CLAUDE_CODE_FALLBACK_VERSION, CODEX_CLIENT_VERSION, type CacheRetention, type ChatErrorReason, type ContentPart, type DoneEvent, type EmbedInputType, type EmbedOptions, type EmbedResult, type EmbedUsage, type ErrorEvent, type ErrorSource, EventStream, FAST_MODE_SUFFIX, type FastModeRejection, type FastModeResolution, type FormattedChatError, type FormattedError, GGAIError, type ImageContent, LOCAL_NO_API_KEY, type Message, type MessageProvenance, type MessageProvenanceKind, type MessageProvenanceSource, type MessageProvenanceVisibility, type OpenRouterProviderPreferences, PRECONNECT_INTERVAL_MS, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type PreconnectOptions, type PreparedContext, type Provider, type ProviderCapability, type ProviderDiagnosticFn, type ProviderEmbedFn, type ProviderEmbedRequest, type ProviderEmbedResponse, type ProviderEntry, ProviderError, type ProviderOptions, type ProviderRerankFn, type ProviderRerankRequest, type ProviderRerankResponse, type ProviderStreamFn, REDACTED as REDACTION_MARKER, type RawContent, type RedactionOptions, type RerankOptions, type RerankResponse, type RerankResult, type RerankUsage, ResponseStalledError, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, TOOL_CALL_NAME_RULES, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolCallNameRule, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, type VideoContent, clampProviderContextImages, classifyProviderError, continuesTurn, describeFastModeRejection, dropInvalidToolCalls, embed, environmentSecrets, formatChatError, formatError, formatErrorForDisplay, getClaudeCliUserAgent, getClaudeCodeVersion, hasLoneSurrogate, isHardBillingMessage, isUsageLimitError, isValidToolCallName, localBearerKey, localWireModelId, opensTurn, palsuAssistantMessage, palsuEmbedding, palsuRerankScore, palsuText, palsuThinking, palsuToolCall, parseFastModeRejection, pendingClaudeCodeVersionRefresh, preconnectOrigin, preconnectProvider, prewarmAnthropicCache, providerRegistry, redactText, redactValue, registerPalsuProvider, rerank, resolveFastModeModel, resolveToolSchema, sanitizeMessagesForWire, setProviderDiagnostic, sliceHead, sliceTail, stream, toAnthropicMessages, toOpenAIMessages, toWellFormedText, toolCallNameRuleFor, usesResponsesLite };
+export { type AssistantMessage, type CacheRetention, type ChatErrorReason, type ContentPart, type DoneEvent, type EmbedInputType, type EmbedOptions, type EmbedResult, type EmbedUsage, type ErrorEvent, type ErrorSource, EventStream, FAST_MODE_SUFFIX, type FastModeRejection, type FastModeResolution, type FormattedChatError, type FormattedError, GGAIError, GG_CODER_USER_AGENT, type ImageContent, LOCAL_NO_API_KEY, type Message, type MessageProvenance, type MessageProvenanceKind, type MessageProvenanceSource, type MessageProvenanceVisibility, type OpenRouterProviderPreferences, PRECONNECT_INTERVAL_MS, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type PreconnectOptions, type PreparedContext, type Provider, type ProviderCapability, type ProviderDiagnosticFn, type ProviderEmbedFn, type ProviderEmbedRequest, type ProviderEmbedResponse, type ProviderEntry, ProviderError, type ProviderOptions, type ProviderRerankFn, type ProviderRerankRequest, type ProviderRerankResponse, type ProviderStreamFn, REDACTED as REDACTION_MARKER, type RawContent, type RedactionOptions, type RerankOptions, type RerankResponse, type RerankResult, type RerankUsage, ResponseStalledError, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, TOOL_CALL_NAME_RULES, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolCallNameRule, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, type VideoContent, clampProviderContextImages, classifyProviderError, continuesTurn, describeFastModeRejection, dropInvalidToolCalls, embed, environmentSecrets, formatChatError, formatError, formatErrorForDisplay, hasLoneSurrogate, isHardBillingMessage, isUsageLimitError, isValidToolCallName, localBearerKey, localWireModelId, opensTurn, palsuAssistantMessage, palsuEmbedding, palsuRerankScore, palsuText, palsuThinking, palsuToolCall, parseFastModeRejection, preconnectOrigin, preconnectProvider, prewarmAnthropicCache, providerRegistry, redactText, redactValue, registerPalsuProvider, rerank, resolveFastModeModel, resolveToolSchema, sanitizeMessagesForWire, setProviderDiagnostic, sliceHead, sliceTail, stream, toAnthropicMessages, toOpenAIMessages, toWellFormedText, toolCallNameRuleFor, usesResponsesLite };

package/dist/index.d.ts

@@ -253,26 +253,18 @@ interface StreamOptions {
     /** OpenAI service tier for latency-sensitive requests. Only sent to first-party OpenAI API calls. */
     serviceTier?: "auto" | "default" | "flex" | "priority";
     /** Provider speed tier. "fast" sends Anthropic `speed: "fast"` (first-party API
-     *  only) or OpenAI/Codex `service_tier: "priority"`. Set automatically by
+     *  only) or OpenAI `service_tier: "priority"`. Set automatically by
      *  `stream()` for fast-mode model ids (see `resolveFastModeModel`). */
     speed?: "standard" | "fast";
-    /** Codex endpoint only: send the Responses-Lite request shape (lite header,
-     *  `parallel_tool_calls: false`, all-turns reasoning context). Unset follows
-     *  the model family (on for gpt-5.6/gpt-6 models). Lite allows only one tool
-     *  call per response, so turning it off lets the model batch calls. */
-    responsesLite?: boolean;
-    /** OpenAI only (Codex and API-key routes): send tools with strict
+    /** OpenAI only (ChatGPT plan and API-key routes): send tools with strict
      *  (grammar-constrained) schemas. Default true. Strict calls carry every
      *  optional field as null, which costs output tokens and latency per call. */
     strictTools?: boolean;
-    /** OpenAI ChatGPT account ID (from OAuth JWT) for codex endpoint */
+    /** ChatGPT plan sign-in: the account subject. Set, it routes OpenAI to the plan. */
     accountId?: string;
-    /** Stable conversation identity for Codex transport headers. This is distinct from
-     *  promptCacheKey: sessions with matching prefixes may share a cache key, but must
-     *  retain independent session/thread identities. */
+    /** Stable conversation identity, distinct from promptCacheKey: sessions with
+     *  matching prefixes may share a cache key but keep independent identities. */
     transportSessionId?: string;
-    /** Google Cloud/Code Assist project ID used by Gemini OAuth transport. */
-    projectId?: string;
     /** Enable provider-native web search. Each provider uses its own format:
      *  - Anthropic: server tool `web_search_20250305`
      *  - Moonshot: `builtin_function` `$web_search`
@@ -315,16 +307,8 @@ interface StreamOptions {
      *  budget_tokens < max_tokens), no request is sent and the response has
      *  zero usage with stopReason "end_turn". */
     prewarm?: boolean;
-    /** Override the User-Agent sent with OAuth-authenticated Anthropic requests.
-     *  Anthropic's OAuth edge rejects requests whose claude-cli version lags too
-     *  far behind the real Claude Code release; callers that track the live
-     *  version should pass it here. Ignored for non-Anthropic providers and for
-     *  Anthropic requests using a regular API key. */
-    userAgent?: string;
-    /** Extra HTTP headers attached to every model request. Used by providers
-     *  whose endpoint gates on client identity (e.g. Kimi For Coding requires a
-     *  `User-Agent: kimi-code-cli/...` and `X-Msh-*` device headers). Merged
-     *  into the underlying SDK's default headers. */
+    /** Extra HTTP headers attached to every model request. Merged into the
+     *  underlying SDK's default headers. */
     defaultHeaders?: Record<string, string>;
     /** Provider-specific request-body fields, merged into the JSON body of every
      *  model request (e.g. OpenRouter's routing object:
@@ -985,8 +969,8 @@ declare const TOOL_CALL_NAME_RULES: Record<ToolCallNameRule["id"], ToolCallNameR
 declare function isValidToolCallName(name: unknown, rule: ToolCallNameRule): boolean;
 /**
  * Pick the tool-name rule for the transport a request will actually use. Mirrors
- * the routing in `stream.ts`: `openai` with an `accountId` is the Codex
- * (Responses) endpoint, MiniMax rides the Anthropic transport but is a third
+ * the routing in `stream.ts`: `openai` itself (the ChatGPT plan or an API key)
+ * is the Responses API, MiniMax rides the Anthropic transport but is a third
  * party, and an `openai` pointed at a custom baseUrl is an unknown compatible
  * server.
  */
@@ -1047,9 +1031,15 @@ declare function continuesTurn(message: Message): boolean;
 /** A real turn boundary: a user message that isn't a runtime notice. */
 declare function opensTurn(message: Message): boolean;
 
-declare const CODEX_CLIENT_VERSION = "0.166.0";
+/**
+ * GPT-6 family (and gpt-5.6): reasoning can't be turned off, and the models
+ * default to low verbosity. Point releases keep the dotted version (gpt-6.1-sol).
+ */
 declare function usesResponsesLite(model: string): boolean;
 
+/** GG Coder's own client identity. Sent where a provider needs a User-Agent; never another vendor's. */
+declare const GG_CODER_USER_AGENT = "ggcoder (+https://github.com/KenKaiii/gg-coder)";
+
 /**
  * Fire a minimal `max_tokens: 1` request that populates the Anthropic prompt
  * cache with the system prompt + tools prefix, so the first real user turn is
@@ -1067,24 +1057,10 @@ declare function prewarmAnthropicCache(options: {
     tools?: StreamOptions["tools"];
     serverTools?: StreamOptions["serverTools"];
     baseUrl?: string;
-    userAgent?: string;
     cacheRetention?: StreamOptions["cacheRetention"];
     signal?: AbortSignal;
 }): Promise<void>;
 
-declare const CLAUDE_CODE_FALLBACK_VERSION = "2.1.316";
-/** Test hook: the pending background refresh, if any. */
-declare function pendingClaudeCodeVersionRefresh(): Promise<void> | null;
-/**
- * Resolve the current Claude Code release version for spoofing the claude-cli
- * User-Agent on OAuth and inference requests. Cached in-memory for the process
- * lifetime and on disk for 24h. Falls back to a hardcoded constant if the npm
- * registry is unreachable and no cache exists.
- */
-declare function getClaudeCodeVersion(): Promise<string>;
-/** Build the User-Agent string Anthropic's OAuth + inference edges expect. */
-declare function getClaudeCliUserAgent(): Promise<string>;
-
 interface PalsuProviderState {
     /** `stream()` calls served. */
     callCount: number;
@@ -1171,4 +1147,4 @@ declare function palsuEmbedding(text: string, dimensions?: number): number[];
 /** Deterministic fake relevance: the share of distinct query words found in the document (0–1). */
 declare function palsuRerankScore(query: string, document: string): number;
 
-export { type AssistantMessage, CLAUDE_CODE_FALLBACK_VERSION, CODEX_CLIENT_VERSION, type CacheRetention, type ChatErrorReason, type ContentPart, type DoneEvent, type EmbedInputType, type EmbedOptions, type EmbedResult, type EmbedUsage, type ErrorEvent, type ErrorSource, EventStream, FAST_MODE_SUFFIX, type FastModeRejection, type FastModeResolution, type FormattedChatError, type FormattedError, GGAIError, type ImageContent, LOCAL_NO_API_KEY, type Message, type MessageProvenance, type MessageProvenanceKind, type MessageProvenanceSource, type MessageProvenanceVisibility, type OpenRouterProviderPreferences, PRECONNECT_INTERVAL_MS, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type PreconnectOptions, type PreparedContext, type Provider, type ProviderCapability, type ProviderDiagnosticFn, type ProviderEmbedFn, type ProviderEmbedRequest, type ProviderEmbedResponse, type ProviderEntry, ProviderError, type ProviderOptions, type ProviderRerankFn, type ProviderRerankRequest, type ProviderRerankResponse, type ProviderStreamFn, REDACTED as REDACTION_MARKER, type RawContent, type RedactionOptions, type RerankOptions, type RerankResponse, type RerankResult, type RerankUsage, ResponseStalledError, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, TOOL_CALL_NAME_RULES, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolCallNameRule, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, type VideoContent, clampProviderContextImages, classifyProviderError, continuesTurn, describeFastModeRejection, dropInvalidToolCalls, embed, environmentSecrets, formatChatError, formatError, formatErrorForDisplay, getClaudeCliUserAgent, getClaudeCodeVersion, hasLoneSurrogate, isHardBillingMessage, isUsageLimitError, isValidToolCallName, localBearerKey, localWireModelId, opensTurn, palsuAssistantMessage, palsuEmbedding, palsuRerankScore, palsuText, palsuThinking, palsuToolCall, parseFastModeRejection, pendingClaudeCodeVersionRefresh, preconnectOrigin, preconnectProvider, prewarmAnthropicCache, providerRegistry, redactText, redactValue, registerPalsuProvider, rerank, resolveFastModeModel, resolveToolSchema, sanitizeMessagesForWire, setProviderDiagnostic, sliceHead, sliceTail, stream, toAnthropicMessages, toOpenAIMessages, toWellFormedText, toolCallNameRuleFor, usesResponsesLite };
+export { type AssistantMessage, type CacheRetention, type ChatErrorReason, type ContentPart, type DoneEvent, type EmbedInputType, type EmbedOptions, type EmbedResult, type EmbedUsage, type ErrorEvent, type ErrorSource, EventStream, FAST_MODE_SUFFIX, type FastModeRejection, type FastModeResolution, type FormattedChatError, type FormattedError, GGAIError, GG_CODER_USER_AGENT, type ImageContent, LOCAL_NO_API_KEY, type Message, type MessageProvenance, type MessageProvenanceKind, type MessageProvenanceSource, type MessageProvenanceVisibility, type OpenRouterProviderPreferences, PRECONNECT_INTERVAL_MS, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type PreconnectOptions, type PreparedContext, type Provider, type ProviderCapability, type ProviderDiagnosticFn, type ProviderEmbedFn, type ProviderEmbedRequest, type ProviderEmbedResponse, type ProviderEntry, ProviderError, type ProviderOptions, type ProviderRerankFn, type ProviderRerankRequest, type ProviderRerankResponse, type ProviderStreamFn, REDACTED as REDACTION_MARKER, type RawContent, type RedactionOptions, type RerankOptions, type RerankResponse, type RerankResult, type RerankUsage, ResponseStalledError, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, TOOL_CALL_NAME_RULES, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolCallNameRule, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, type VideoContent, clampProviderContextImages, classifyProviderError, continuesTurn, describeFastModeRejection, dropInvalidToolCalls, embed, environmentSecrets, formatChatError, formatError, formatErrorForDisplay, hasLoneSurrogate, isHardBillingMessage, isUsageLimitError, isValidToolCallName, localBearerKey, localWireModelId, opensTurn, palsuAssistantMessage, palsuEmbedding, palsuRerankScore, palsuText, palsuThinking, palsuToolCall, parseFastModeRejection, preconnectOrigin, preconnectProvider, prewarmAnthropicCache, providerRegistry, redactText, redactValue, registerPalsuProvider, rerank, resolveFastModeModel, resolveToolSchema, sanitizeMessagesForWire, setProviderDiagnostic, sliceHead, sliceTail, stream, toAnthropicMessages, toOpenAIMessages, toWellFormedText, toolCallNameRuleFor, usesResponsesLite };