@mariozechner/pi-ai 0.70.6 → 0.71.1

package/dist/types.js.map

@@ -1 +1 @@
-{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"","sourcesContent":["import type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type KnownApi =\n\t| \"openai-completions\"\n\t| \"mistral-conversations\"\n\t| \"openai-responses\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex-responses\"\n\t| \"anthropic-messages\"\n\t| \"bedrock-converse-stream\"\n\t| \"google-generative-ai\"\n\t| \"google-gemini-cli\"\n\t| \"google-vertex\";\n\nexport type Api = KnownApi | (string & {});\n\nexport type KnownProvider =\n\t| \"amazon-bedrock\"\n\t| \"anthropic\"\n\t| \"google\"\n\t| \"google-gemini-cli\"\n\t| \"google-antigravity\"\n\t| \"google-vertex\"\n\t| \"openai\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex\"\n\t| \"deepseek\"\n\t| \"github-copilot\"\n\t| \"xai\"\n\t| \"groq\"\n\t| \"cerebras\"\n\t| \"openrouter\"\n\t| \"vercel-ai-gateway\"\n\t| \"zai\"\n\t| \"mistral\"\n\t| \"minimax\"\n\t| \"minimax-cn\"\n\t| \"huggingface\"\n\t| \"fireworks\"\n\t| \"opencode\"\n\t| \"opencode-go\"\n\t| \"kimi-coding\"\n\t| \"cloudflare-workers-ai\";\nexport type Provider = KnownProvider | string;\n\nexport type ThinkingLevel = \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\";\n\n/** Token budgets for each thinking level (token-based providers only) */\nexport interface ThinkingBudgets {\n\tminimal?: number;\n\tlow?: number;\n\tmedium?: number;\n\thigh?: number;\n}\n\n// Base options all providers share\nexport type CacheRetention = \"none\" | \"short\" | \"long\";\n\nexport type Transport = \"sse\" | \"websocket\" | \"auto\";\n\nexport interface ProviderResponse {\n\tstatus: number;\n\theaders: Record<string, string>;\n}\n\nexport interface StreamOptions {\n\ttemperature?: number;\n\tmaxTokens?: number;\n\tsignal?: AbortSignal;\n\tapiKey?: string;\n\t/**\n\t * Preferred transport for providers that support multiple transports.\n\t * Providers that do not support this option ignore it.\n\t */\n\ttransport?: Transport;\n\t/**\n\t * Prompt cache retention preference. Providers map this to their supported values.\n\t * Default: \"short\".\n\t */\n\tcacheRetention?: CacheRetention;\n\t/**\n\t * Optional session identifier for providers that support session-based caching.\n\t * Providers can use this to enable prompt caching, request routing, or other\n\t * session-aware features. Ignored by providers that don't support it.\n\t */\n\tsessionId?: string;\n\t/**\n\t * Optional callback for inspecting or replacing provider payloads before sending.\n\t * Return undefined to keep the payload unchanged.\n\t */\n\tonPayload?: (payload: unknown, model: Model<Api>) => unknown | undefined | Promise<unknown | undefined>;\n\t/**\n\t * Optional callback invoked after an HTTP response is received and before\n\t * its body stream is consumed.\n\t */\n\tonResponse?: (response: ProviderResponse, model: Model<Api>) => void | Promise<void>;\n\t/**\n\t * Optional custom HTTP headers to include in API requests.\n\t * Merged with provider defaults; can override default headers.\n\t * Not supported by all providers (e.g., AWS Bedrock uses SDK auth).\n\t */\n\theaders?: Record<string, string>;\n\t/**\n\t * HTTP request timeout in milliseconds for providers/SDKs that support it.\n\t * For example, OpenAI and Anthropic SDK clients default to 10 minutes.\n\t */\n\ttimeoutMs?: number;\n\t/**\n\t * Maximum retry attempts for providers/SDKs that support client-side retries.\n\t * For example, OpenAI and Anthropic SDK clients default to 2.\n\t */\n\tmaxRetries?: number;\n\t/**\n\t * Maximum delay in milliseconds to wait for a retry when the server requests a long wait.\n\t * If the server's requested delay exceeds this value, the request fails immediately\n\t * with an error containing the requested delay, allowing higher-level retry logic\n\t * to handle it with user visibility.\n\t * Default: 60000 (60 seconds). Set to 0 to disable the cap.\n\t */\n\tmaxRetryDelayMs?: number;\n\t/**\n\t * Optional metadata to include in API requests.\n\t * Providers extract the fields they understand and ignore the rest.\n\t * For example, Anthropic uses `user_id` for abuse tracking and rate limiting.\n\t */\n\tmetadata?: Record<string, unknown>;\n}\n\nexport type ProviderStreamOptions = StreamOptions & Record<string, unknown>;\n\n// Unified options with reasoning passed to streamSimple() and completeSimple()\nexport interface SimpleStreamOptions extends StreamOptions {\n\treasoning?: ThinkingLevel;\n\t/** Custom token budgets for thinking levels (token-based providers only) */\n\tthinkingBudgets?: ThinkingBudgets;\n}\n\n// Generic StreamFunction with typed options.\n//\n// Contract:\n// - Must return an AssistantMessageEventStream.\n// - Once invoked, request/model/runtime failures should be encoded in the\n//   returned stream, not thrown.\n// - Error termination must produce an AssistantMessage with stopReason\n//   \"error\" or \"aborted\" and errorMessage, emitted via the stream protocol.\nexport type StreamFunction<TApi extends Api = Api, TOptions extends StreamOptions = StreamOptions> = (\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: TOptions,\n) => AssistantMessageEventStream;\n\nexport interface TextSignatureV1 {\n\tv: 1;\n\tid: string;\n\tphase?: \"commentary\" | \"final_answer\";\n}\n\nexport interface TextContent {\n\ttype: \"text\";\n\ttext: string;\n\ttextSignature?: string; // e.g., for OpenAI responses, message metadata (legacy id string or TextSignatureV1 JSON)\n}\n\nexport interface ThinkingContent {\n\ttype: \"thinking\";\n\tthinking: string;\n\tthinkingSignature?: string; // e.g., for OpenAI responses, the reasoning item ID\n\t/** When true, the thinking content was redacted by safety filters. The opaque\n\t *  encrypted payload is stored in `thinkingSignature` so it can be passed back\n\t *  to the API for multi-turn continuity. */\n\tredacted?: boolean;\n}\n\nexport interface ImageContent {\n\ttype: \"image\";\n\tdata: string; // base64 encoded image data\n\tmimeType: string; // e.g., \"image/jpeg\", \"image/png\"\n}\n\nexport interface ToolCall {\n\ttype: \"toolCall\";\n\tid: string;\n\tname: string;\n\targuments: Record<string, any>;\n\tthoughtSignature?: string; // Google-specific: opaque signature for reusing thought context\n}\n\nexport interface Usage {\n\tinput: number;\n\toutput: number;\n\tcacheRead: number;\n\tcacheWrite: number;\n\ttotalTokens: number;\n\tcost: {\n\t\tinput: number;\n\t\toutput: number;\n\t\tcacheRead: number;\n\t\tcacheWrite: number;\n\t\ttotal: number;\n\t};\n}\n\nexport type StopReason = \"stop\" | \"length\" | \"toolUse\" | \"error\" | \"aborted\";\n\nexport interface UserMessage {\n\trole: \"user\";\n\tcontent: string | (TextContent | ImageContent)[];\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface AssistantMessage {\n\trole: \"assistant\";\n\tcontent: (TextContent | ThinkingContent | ToolCall)[];\n\tapi: Api;\n\tprovider: Provider;\n\tmodel: string;\n\tresponseId?: string; // Provider-specific response/message identifier when the upstream API exposes one\n\tusage: Usage;\n\tstopReason: StopReason;\n\terrorMessage?: string;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface ToolResultMessage<TDetails = any> {\n\trole: \"toolResult\";\n\ttoolCallId: string;\n\ttoolName: string;\n\tcontent: (TextContent | ImageContent)[]; // Supports text and images\n\tdetails?: TDetails;\n\tisError: boolean;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport type Message = UserMessage | AssistantMessage | ToolResultMessage;\n\nimport type { TSchema } from \"typebox\";\n\nexport interface Tool<TParameters extends TSchema = TSchema> {\n\tname: string;\n\tdescription: string;\n\tparameters: TParameters;\n}\n\nexport interface Context {\n\tsystemPrompt?: string;\n\tmessages: Message[];\n\ttools?: Tool[];\n}\n\n/**\n * Event protocol for AssistantMessageEventStream.\n *\n * Streams should emit `start` before partial updates, then terminate with either:\n * - `done` carrying the final successful AssistantMessage, or\n * - `error` carrying the final AssistantMessage with stopReason \"error\" or \"aborted\"\n *   and errorMessage.\n */\nexport type AssistantMessageEvent =\n\t| { type: \"start\"; partial: AssistantMessage }\n\t| { type: \"text_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"text_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"text_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"thinking_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"thinking_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"thinking_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"toolcall_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"toolcall_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"toolcall_end\"; contentIndex: number; toolCall: ToolCall; partial: AssistantMessage }\n\t| { type: \"done\"; reason: Extract<StopReason, \"stop\" | \"length\" | \"toolUse\">; message: AssistantMessage }\n\t| { type: \"error\"; reason: Extract<StopReason, \"aborted\" | \"error\">; error: AssistantMessage };\n\n/**\n * Compatibility settings for OpenAI-compatible completions APIs.\n * Use this to override URL-based auto-detection for custom providers.\n */\nexport interface OpenAICompletionsCompat {\n\t/** Whether the provider supports the `store` field. Default: auto-detected from URL. */\n\tsupportsStore?: boolean;\n\t/** Whether the provider supports the `developer` role (vs `system`). Default: auto-detected from URL. */\n\tsupportsDeveloperRole?: boolean;\n\t/** Whether the provider supports `reasoning_effort`. Default: auto-detected from URL. */\n\tsupportsReasoningEffort?: boolean;\n\t/** Optional mapping from pi-ai reasoning levels to provider/model-specific `reasoning_effort` values. */\n\treasoningEffortMap?: Partial<Record<ThinkingLevel, string>>;\n\t/** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */\n\tsupportsUsageInStreaming?: boolean;\n\t/** Which field to use for max tokens. Default: auto-detected from URL. */\n\tmaxTokensField?: \"max_completion_tokens\" | \"max_tokens\";\n\t/** Whether tool results require the `name` field. Default: auto-detected from URL. */\n\trequiresToolResultName?: boolean;\n\t/** Whether a user message after tool results requires an assistant message in between. Default: auto-detected from URL. */\n\trequiresAssistantAfterToolResult?: boolean;\n\t/** Whether thinking blocks must be converted to text blocks with <thinking> delimiters. Default: auto-detected from URL. */\n\trequiresThinkingAsText?: boolean;\n\t/** Whether all replayed assistant messages must include an empty reasoning_content field when reasoning is enabled. Default: auto-detected from URL. */\n\trequiresReasoningContentOnAssistantMessages?: boolean;\n\t/** Format for reasoning/thinking parameter. \"openai\" uses reasoning_effort, \"openrouter\" uses reasoning: { effort }, \"deepseek\" uses thinking: { type } plus reasoning_effort, \"zai\" uses top-level enable_thinking: boolean, \"qwen\" uses top-level enable_thinking: boolean, and \"qwen-chat-template\" uses chat_template_kwargs.enable_thinking. Default: \"openai\". */\n\tthinkingFormat?: \"openai\" | \"openrouter\" | \"deepseek\" | \"zai\" | \"qwen\" | \"qwen-chat-template\";\n\t/** OpenRouter-specific routing preferences. Only used when baseUrl points to OpenRouter. */\n\topenRouterRouting?: OpenRouterRouting;\n\t/** Vercel AI Gateway routing preferences. Only used when baseUrl points to Vercel AI Gateway. */\n\tvercelGatewayRouting?: VercelGatewayRouting;\n\t/** Whether z.ai supports top-level `tool_stream: true` for streaming tool call deltas. Default: false. */\n\tzaiToolStream?: boolean;\n\t/** Whether the provider supports the `strict` field in tool definitions. Default: true. */\n\tsupportsStrictMode?: boolean;\n\t/** Cache control convention for prompt caching. \"anthropic\" applies Anthropic-style `cache_control` markers to the system prompt, last tool definition, and last user/assistant text content. */\n\tcacheControlFormat?: \"anthropic\";\n\t/** Whether to send known session-affinity headers (`session_id`, `x-client-request-id`, `x-session-affinity`) from `options.sessionId` when caching is enabled. Default: false. */\n\tsendSessionAffinityHeaders?: boolean;\n\t/** Whether the provider supports long prompt cache retention (`prompt_cache_retention: \"24h\"` or Anthropic-style `cache_control.ttl: \"1h\"`, depending on format). Default: true. */\n\tsupportsLongCacheRetention?: boolean;\n}\n\n/** Compatibility settings for OpenAI Responses APIs. */\nexport interface OpenAIResponsesCompat {\n\t/** Whether to send the OpenAI `session_id` cache-affinity header from `options.sessionId` when caching is enabled. Default: true. */\n\tsendSessionIdHeader?: boolean;\n\t/** Whether the provider supports `prompt_cache_retention: \"24h\"`. Default: true. */\n\tsupportsLongCacheRetention?: boolean;\n}\n\n/** Compatibility settings for Anthropic Messages-compatible APIs. */\nexport interface AnthropicMessagesCompat {\n\t/**\n\t * Whether the provider accepts per-tool `eager_input_streaming`.\n\t * When false, the Anthropic provider omits `tools[].eager_input_streaming`\n\t * and sends the legacy `fine-grained-tool-streaming-2025-05-14` beta header\n\t * for tool-enabled requests.\n\t * Default: true.\n\t */\n\tsupportsEagerToolInputStreaming?: boolean;\n\t/** Whether the provider supports Anthropic long cache retention (`cache_control.ttl: \"1h\"`). Default: true. */\n\tsupportsLongCacheRetention?: boolean;\n}\n\n/**\n * OpenRouter provider routing preferences.\n * Controls which upstream providers OpenRouter routes requests to.\n * Sent as the `provider` field in the OpenRouter API request body.\n * @see https://openrouter.ai/docs/guides/routing/provider-selection\n */\nexport interface OpenRouterRouting {\n\t/** Whether to allow backup providers to serve requests. Default: true. */\n\tallow_fallbacks?: boolean;\n\t/** Whether to filter providers to only those that support all parameters in the request. Default: false. */\n\trequire_parameters?: boolean;\n\t/** Data collection setting. \"allow\" (default): allow providers that may store/train on data. \"deny\": only use providers that don't collect user data. */\n\tdata_collection?: \"deny\" | \"allow\";\n\t/** Whether to restrict routing to only ZDR (Zero Data Retention) endpoints. */\n\tzdr?: boolean;\n\t/** Whether to restrict routing to only models that allow text distillation. */\n\tenforce_distillable_text?: boolean;\n\t/** An ordered list of provider names/slugs to try in sequence, falling back to the next if unavailable. */\n\torder?: string[];\n\t/** List of provider names/slugs to exclusively allow for this request. */\n\tonly?: string[];\n\t/** List of provider names/slugs to skip for this request. */\n\tignore?: string[];\n\t/** A list of quantization levels to filter providers by (e.g., [\"fp16\", \"bf16\", \"fp8\", \"fp6\", \"int8\", \"int4\", \"fp4\", \"fp32\"]). */\n\tquantizations?: string[];\n\t/** Sorting strategy. Can be a string (e.g., \"price\", \"throughput\", \"latency\") or an object with `by` and `partition`. */\n\tsort?:\n\t\t| string\n\t\t| {\n\t\t\t\t/** The sorting metric: \"price\", \"throughput\", \"latency\". */\n\t\t\t\tby?: string;\n\t\t\t\t/** Partitioning strategy: \"model\" (default) or \"none\". */\n\t\t\t\tpartition?: string | null;\n\t\t  };\n\t/** Maximum price per million tokens (USD). */\n\tmax_price?: {\n\t\t/** Price per million prompt tokens. */\n\t\tprompt?: number | string;\n\t\t/** Price per million completion tokens. */\n\t\tcompletion?: number | string;\n\t\t/** Price per image. */\n\t\timage?: number | string;\n\t\t/** Price per audio unit. */\n\t\taudio?: number | string;\n\t\t/** Price per request. */\n\t\trequest?: number | string;\n\t};\n\t/** Preferred minimum throughput (tokens/second). Can be a number (applies to p50) or an object with percentile-specific cutoffs. */\n\tpreferred_min_throughput?:\n\t\t| number\n\t\t| {\n\t\t\t\t/** Minimum tokens/second at the 50th percentile. */\n\t\t\t\tp50?: number;\n\t\t\t\t/** Minimum tokens/second at the 75th percentile. */\n\t\t\t\tp75?: number;\n\t\t\t\t/** Minimum tokens/second at the 90th percentile. */\n\t\t\t\tp90?: number;\n\t\t\t\t/** Minimum tokens/second at the 99th percentile. */\n\t\t\t\tp99?: number;\n\t\t  };\n\t/** Preferred maximum latency (seconds). Can be a number (applies to p50) or an object with percentile-specific cutoffs. */\n\tpreferred_max_latency?:\n\t\t| number\n\t\t| {\n\t\t\t\t/** Maximum latency in seconds at the 50th percentile. */\n\t\t\t\tp50?: number;\n\t\t\t\t/** Maximum latency in seconds at the 75th percentile. */\n\t\t\t\tp75?: number;\n\t\t\t\t/** Maximum latency in seconds at the 90th percentile. */\n\t\t\t\tp90?: number;\n\t\t\t\t/** Maximum latency in seconds at the 99th percentile. */\n\t\t\t\tp99?: number;\n\t\t  };\n}\n\n/**\n * Vercel AI Gateway routing preferences.\n * Controls which upstream providers the gateway routes requests to.\n * @see https://vercel.com/docs/ai-gateway/models-and-providers/provider-options\n */\nexport interface VercelGatewayRouting {\n\t/** List of provider slugs to exclusively use for this request (e.g., [\"bedrock\", \"anthropic\"]). */\n\tonly?: string[];\n\t/** List of provider slugs to try in order (e.g., [\"anthropic\", \"openai\"]). */\n\torder?: string[];\n}\n\n// Model interface for the unified model system\nexport interface Model<TApi extends Api> {\n\tid: string;\n\tname: string;\n\tapi: TApi;\n\tprovider: Provider;\n\tbaseUrl: string;\n\treasoning: boolean;\n\tinput: (\"text\" | \"image\")[];\n\tcost: {\n\t\tinput: number; // $/million tokens\n\t\toutput: number; // $/million tokens\n\t\tcacheRead: number; // $/million tokens\n\t\tcacheWrite: number; // $/million tokens\n\t};\n\tcontextWindow: number;\n\tmaxTokens: number;\n\theaders?: Record<string, string>;\n\t/** Compatibility overrides for OpenAI-compatible APIs. If not set, auto-detected from baseUrl. */\n\tcompat?: TApi extends \"openai-completions\"\n\t\t? OpenAICompletionsCompat\n\t\t: TApi extends \"openai-responses\"\n\t\t\t? OpenAIResponsesCompat\n\t\t\t: TApi extends \"anthropic-messages\"\n\t\t\t\t? AnthropicMessagesCompat\n\t\t\t\t: never;\n}\n"]}
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"","sourcesContent":["import type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type KnownApi =\n\t| \"openai-completions\"\n\t| \"mistral-conversations\"\n\t| \"openai-responses\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex-responses\"\n\t| \"anthropic-messages\"\n\t| \"bedrock-converse-stream\"\n\t| \"google-generative-ai\"\n\t| \"google-vertex\";\n\nexport type Api = KnownApi | (string & {});\n\nexport type KnownProvider =\n\t| \"amazon-bedrock\"\n\t| \"anthropic\"\n\t| \"google\"\n\t| \"google-vertex\"\n\t| \"openai\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex\"\n\t| \"deepseek\"\n\t| \"github-copilot\"\n\t| \"xai\"\n\t| \"groq\"\n\t| \"cerebras\"\n\t| \"openrouter\"\n\t| \"vercel-ai-gateway\"\n\t| \"zai\"\n\t| \"mistral\"\n\t| \"minimax\"\n\t| \"minimax-cn\"\n\t| \"moonshotai\"\n\t| \"moonshotai-cn\"\n\t| \"huggingface\"\n\t| \"fireworks\"\n\t| \"opencode\"\n\t| \"opencode-go\"\n\t| \"kimi-coding\"\n\t| \"cloudflare-workers-ai\"\n\t| \"cloudflare-ai-gateway\";\nexport type Provider = KnownProvider | string;\n\nexport type ThinkingLevel = \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\";\n\n/** Token budgets for each thinking level (token-based providers only) */\nexport interface ThinkingBudgets {\n\tminimal?: number;\n\tlow?: number;\n\tmedium?: number;\n\thigh?: number;\n}\n\n// Base options all providers share\nexport type CacheRetention = \"none\" | \"short\" | \"long\";\n\nexport type Transport = \"sse\" | \"websocket\" | \"websocket-cached\" | \"auto\";\n\nexport interface ProviderResponse {\n\tstatus: number;\n\theaders: Record<string, string>;\n}\n\nexport interface StreamOptions {\n\ttemperature?: number;\n\tmaxTokens?: number;\n\tsignal?: AbortSignal;\n\tapiKey?: string;\n\t/**\n\t * Preferred transport for providers that support multiple transports.\n\t * Providers that do not support this option ignore it.\n\t */\n\ttransport?: Transport;\n\t/**\n\t * Prompt cache retention preference. Providers map this to their supported values.\n\t * Default: \"short\".\n\t */\n\tcacheRetention?: CacheRetention;\n\t/**\n\t * Optional session identifier for providers that support session-based caching.\n\t * Providers can use this to enable prompt caching, request routing, or other\n\t * session-aware features. Ignored by providers that don't support it.\n\t */\n\tsessionId?: string;\n\t/**\n\t * Optional callback for inspecting or replacing provider payloads before sending.\n\t * Return undefined to keep the payload unchanged.\n\t */\n\tonPayload?: (payload: unknown, model: Model<Api>) => unknown | undefined | Promise<unknown | undefined>;\n\t/**\n\t * Optional callback invoked after an HTTP response is received and before\n\t * its body stream is consumed.\n\t */\n\tonResponse?: (response: ProviderResponse, model: Model<Api>) => void | Promise<void>;\n\t/**\n\t * Optional custom HTTP headers to include in API requests.\n\t * Merged with provider defaults; can override default headers.\n\t * Not supported by all providers (e.g., AWS Bedrock uses SDK auth).\n\t */\n\theaders?: Record<string, string>;\n\t/**\n\t * HTTP request timeout in milliseconds for providers/SDKs that support it.\n\t * For example, OpenAI and Anthropic SDK clients default to 10 minutes.\n\t */\n\ttimeoutMs?: number;\n\t/**\n\t * Maximum retry attempts for providers/SDKs that support client-side retries.\n\t * For example, OpenAI and Anthropic SDK clients default to 2.\n\t */\n\tmaxRetries?: number;\n\t/**\n\t * Maximum delay in milliseconds to wait for a retry when the server requests a long wait.\n\t * If the server's requested delay exceeds this value, the request fails immediately\n\t * with an error containing the requested delay, allowing higher-level retry logic\n\t * to handle it with user visibility.\n\t * Default: 60000 (60 seconds). Set to 0 to disable the cap.\n\t */\n\tmaxRetryDelayMs?: number;\n\t/**\n\t * Optional metadata to include in API requests.\n\t * Providers extract the fields they understand and ignore the rest.\n\t * For example, Anthropic uses `user_id` for abuse tracking and rate limiting.\n\t */\n\tmetadata?: Record<string, unknown>;\n}\n\nexport type ProviderStreamOptions = StreamOptions & Record<string, unknown>;\n\n// Unified options with reasoning passed to streamSimple() and completeSimple()\nexport interface SimpleStreamOptions extends StreamOptions {\n\treasoning?: ThinkingLevel;\n\t/** Custom token budgets for thinking levels (token-based providers only) */\n\tthinkingBudgets?: ThinkingBudgets;\n}\n\n// Generic StreamFunction with typed options.\n//\n// Contract:\n// - Must return an AssistantMessageEventStream.\n// - Once invoked, request/model/runtime failures should be encoded in the\n//   returned stream, not thrown.\n// - Error termination must produce an AssistantMessage with stopReason\n//   \"error\" or \"aborted\" and errorMessage, emitted via the stream protocol.\nexport type StreamFunction<TApi extends Api = Api, TOptions extends StreamOptions = StreamOptions> = (\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: TOptions,\n) => AssistantMessageEventStream;\n\nexport interface TextSignatureV1 {\n\tv: 1;\n\tid: string;\n\tphase?: \"commentary\" | \"final_answer\";\n}\n\nexport interface TextContent {\n\ttype: \"text\";\n\ttext: string;\n\ttextSignature?: string; // e.g., for OpenAI responses, message metadata (legacy id string or TextSignatureV1 JSON)\n}\n\nexport interface ThinkingContent {\n\ttype: \"thinking\";\n\tthinking: string;\n\tthinkingSignature?: string; // e.g., for OpenAI responses, the reasoning item ID\n\t/** When true, the thinking content was redacted by safety filters. The opaque\n\t *  encrypted payload is stored in `thinkingSignature` so it can be passed back\n\t *  to the API for multi-turn continuity. */\n\tredacted?: boolean;\n}\n\nexport interface ImageContent {\n\ttype: \"image\";\n\tdata: string; // base64 encoded image data\n\tmimeType: string; // e.g., \"image/jpeg\", \"image/png\"\n}\n\nexport interface ToolCall {\n\ttype: \"toolCall\";\n\tid: string;\n\tname: string;\n\targuments: Record<string, any>;\n\tthoughtSignature?: string; // Google-specific: opaque signature for reusing thought context\n}\n\nexport interface Usage {\n\tinput: number;\n\toutput: number;\n\tcacheRead: number;\n\tcacheWrite: number;\n\ttotalTokens: number;\n\tcost: {\n\t\tinput: number;\n\t\toutput: number;\n\t\tcacheRead: number;\n\t\tcacheWrite: number;\n\t\ttotal: number;\n\t};\n}\n\nexport type StopReason = \"stop\" | \"length\" | \"toolUse\" | \"error\" | \"aborted\";\n\nexport interface UserMessage {\n\trole: \"user\";\n\tcontent: string | (TextContent | ImageContent)[];\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface AssistantMessage {\n\trole: \"assistant\";\n\tcontent: (TextContent | ThinkingContent | ToolCall)[];\n\tapi: Api;\n\tprovider: Provider;\n\tmodel: string;\n\tresponseModel?: string; // Concrete `chunk.model` when different from the requested `model` (e.g. OpenRouter `auto` -> `anthropic/...`)\n\tresponseId?: string; // Provider-specific response/message identifier when the upstream API exposes one\n\tusage: Usage;\n\tstopReason: StopReason;\n\terrorMessage?: string;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface ToolResultMessage<TDetails = any> {\n\trole: \"toolResult\";\n\ttoolCallId: string;\n\ttoolName: string;\n\tcontent: (TextContent | ImageContent)[]; // Supports text and images\n\tdetails?: TDetails;\n\tisError: boolean;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport type Message = UserMessage | AssistantMessage | ToolResultMessage;\n\nimport type { TSchema } from \"typebox\";\n\nexport interface Tool<TParameters extends TSchema = TSchema> {\n\tname: string;\n\tdescription: string;\n\tparameters: TParameters;\n}\n\nexport interface Context {\n\tsystemPrompt?: string;\n\tmessages: Message[];\n\ttools?: Tool[];\n}\n\n/**\n * Event protocol for AssistantMessageEventStream.\n *\n * Streams should emit `start` before partial updates, then terminate with either:\n * - `done` carrying the final successful AssistantMessage, or\n * - `error` carrying the final AssistantMessage with stopReason \"error\" or \"aborted\"\n *   and errorMessage.\n */\nexport type AssistantMessageEvent =\n\t| { type: \"start\"; partial: AssistantMessage }\n\t| { type: \"text_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"text_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"text_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"thinking_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"thinking_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"thinking_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"toolcall_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"toolcall_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"toolcall_end\"; contentIndex: number; toolCall: ToolCall; partial: AssistantMessage }\n\t| { type: \"done\"; reason: Extract<StopReason, \"stop\" | \"length\" | \"toolUse\">; message: AssistantMessage }\n\t| { type: \"error\"; reason: Extract<StopReason, \"aborted\" | \"error\">; error: AssistantMessage };\n\n/**\n * Compatibility settings for OpenAI-compatible completions APIs.\n * Use this to override URL-based auto-detection for custom providers.\n */\nexport interface OpenAICompletionsCompat {\n\t/** Whether the provider supports the `store` field. Default: auto-detected from URL. */\n\tsupportsStore?: boolean;\n\t/** Whether the provider supports the `developer` role (vs `system`). Default: auto-detected from URL. */\n\tsupportsDeveloperRole?: boolean;\n\t/** Whether the provider supports `reasoning_effort`. Default: auto-detected from URL. */\n\tsupportsReasoningEffort?: boolean;\n\t/** Optional mapping from pi-ai reasoning levels to provider/model-specific `reasoning_effort` values. */\n\treasoningEffortMap?: Partial<Record<ThinkingLevel, string>>;\n\t/** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */\n\tsupportsUsageInStreaming?: boolean;\n\t/** Which field to use for max tokens. Default: auto-detected from URL. */\n\tmaxTokensField?: \"max_completion_tokens\" | \"max_tokens\";\n\t/** Whether tool results require the `name` field. Default: auto-detected from URL. */\n\trequiresToolResultName?: boolean;\n\t/** Whether a user message after tool results requires an assistant message in between. Default: auto-detected from URL. */\n\trequiresAssistantAfterToolResult?: boolean;\n\t/** Whether thinking blocks must be converted to text blocks with <thinking> delimiters. Default: auto-detected from URL. */\n\trequiresThinkingAsText?: boolean;\n\t/** Whether all replayed assistant messages must include an empty reasoning_content field when reasoning is enabled. Default: auto-detected from URL. */\n\trequiresReasoningContentOnAssistantMessages?: boolean;\n\t/** Format for reasoning/thinking parameter. \"openai\" uses reasoning_effort, \"openrouter\" uses reasoning: { effort }, \"deepseek\" uses thinking: { type } plus reasoning_effort, \"zai\" uses top-level enable_thinking: boolean, \"qwen\" uses top-level enable_thinking: boolean, and \"qwen-chat-template\" uses chat_template_kwargs.enable_thinking. Default: \"openai\". */\n\tthinkingFormat?: \"openai\" | \"openrouter\" | \"deepseek\" | \"zai\" | \"qwen\" | \"qwen-chat-template\";\n\t/** OpenRouter-specific routing preferences. Only used when baseUrl points to OpenRouter. */\n\topenRouterRouting?: OpenRouterRouting;\n\t/** Vercel AI Gateway routing preferences. Only used when baseUrl points to Vercel AI Gateway. */\n\tvercelGatewayRouting?: VercelGatewayRouting;\n\t/** Whether z.ai supports top-level `tool_stream: true` for streaming tool call deltas. Default: false. */\n\tzaiToolStream?: boolean;\n\t/** Whether the provider supports the `strict` field in tool definitions. Default: true. */\n\tsupportsStrictMode?: boolean;\n\t/** Cache control convention for prompt caching. \"anthropic\" applies Anthropic-style `cache_control` markers to the system prompt, last tool definition, and last user/assistant text content. */\n\tcacheControlFormat?: \"anthropic\";\n\t/** Whether to send known session-affinity headers (`session_id`, `x-client-request-id`, `x-session-affinity`) from `options.sessionId` when caching is enabled. Default: false. */\n\tsendSessionAffinityHeaders?: boolean;\n\t/** Whether the provider supports long prompt cache retention (`prompt_cache_retention: \"24h\"` or Anthropic-style `cache_control.ttl: \"1h\"`, depending on format). Default: true. */\n\tsupportsLongCacheRetention?: boolean;\n}\n\n/** Compatibility settings for OpenAI Responses APIs. */\nexport interface OpenAIResponsesCompat {\n\t/** Whether to send the OpenAI `session_id` cache-affinity header from `options.sessionId` when caching is enabled. Default: true. */\n\tsendSessionIdHeader?: boolean;\n\t/** Whether the provider supports `prompt_cache_retention: \"24h\"`. Default: true. */\n\tsupportsLongCacheRetention?: boolean;\n}\n\n/** Compatibility settings for Anthropic Messages-compatible APIs. */\nexport interface AnthropicMessagesCompat {\n\t/**\n\t * Whether the provider accepts per-tool `eager_input_streaming`.\n\t * When false, the Anthropic provider omits `tools[].eager_input_streaming`\n\t * and sends the legacy `fine-grained-tool-streaming-2025-05-14` beta header\n\t * for tool-enabled requests.\n\t * Default: true.\n\t */\n\tsupportsEagerToolInputStreaming?: boolean;\n\t/** Whether the provider supports Anthropic long cache retention (`cache_control.ttl: \"1h\"`). Default: true. */\n\tsupportsLongCacheRetention?: boolean;\n}\n\n/**\n * OpenRouter provider routing preferences.\n * Controls which upstream providers OpenRouter routes requests to.\n * Sent as the `provider` field in the OpenRouter API request body.\n * @see https://openrouter.ai/docs/guides/routing/provider-selection\n */\nexport interface OpenRouterRouting {\n\t/** Whether to allow backup providers to serve requests. Default: true. */\n\tallow_fallbacks?: boolean;\n\t/** Whether to filter providers to only those that support all parameters in the request. Default: false. */\n\trequire_parameters?: boolean;\n\t/** Data collection setting. \"allow\" (default): allow providers that may store/train on data. \"deny\": only use providers that don't collect user data. */\n\tdata_collection?: \"deny\" | \"allow\";\n\t/** Whether to restrict routing to only ZDR (Zero Data Retention) endpoints. */\n\tzdr?: boolean;\n\t/** Whether to restrict routing to only models that allow text distillation. */\n\tenforce_distillable_text?: boolean;\n\t/** An ordered list of provider names/slugs to try in sequence, falling back to the next if unavailable. */\n\torder?: string[];\n\t/** List of provider names/slugs to exclusively allow for this request. */\n\tonly?: string[];\n\t/** List of provider names/slugs to skip for this request. */\n\tignore?: string[];\n\t/** A list of quantization levels to filter providers by (e.g., [\"fp16\", \"bf16\", \"fp8\", \"fp6\", \"int8\", \"int4\", \"fp4\", \"fp32\"]). */\n\tquantizations?: string[];\n\t/** Sorting strategy. Can be a string (e.g., \"price\", \"throughput\", \"latency\") or an object with `by` and `partition`. */\n\tsort?:\n\t\t| string\n\t\t| {\n\t\t\t\t/** The sorting metric: \"price\", \"throughput\", \"latency\". */\n\t\t\t\tby?: string;\n\t\t\t\t/** Partitioning strategy: \"model\" (default) or \"none\". */\n\t\t\t\tpartition?: string | null;\n\t\t  };\n\t/** Maximum price per million tokens (USD). */\n\tmax_price?: {\n\t\t/** Price per million prompt tokens. */\n\t\tprompt?: number | string;\n\t\t/** Price per million completion tokens. */\n\t\tcompletion?: number | string;\n\t\t/** Price per image. */\n\t\timage?: number | string;\n\t\t/** Price per audio unit. */\n\t\taudio?: number | string;\n\t\t/** Price per request. */\n\t\trequest?: number | string;\n\t};\n\t/** Preferred minimum throughput (tokens/second). Can be a number (applies to p50) or an object with percentile-specific cutoffs. */\n\tpreferred_min_throughput?:\n\t\t| number\n\t\t| {\n\t\t\t\t/** Minimum tokens/second at the 50th percentile. */\n\t\t\t\tp50?: number;\n\t\t\t\t/** Minimum tokens/second at the 75th percentile. */\n\t\t\t\tp75?: number;\n\t\t\t\t/** Minimum tokens/second at the 90th percentile. */\n\t\t\t\tp90?: number;\n\t\t\t\t/** Minimum tokens/second at the 99th percentile. */\n\t\t\t\tp99?: number;\n\t\t  };\n\t/** Preferred maximum latency (seconds). Can be a number (applies to p50) or an object with percentile-specific cutoffs. */\n\tpreferred_max_latency?:\n\t\t| number\n\t\t| {\n\t\t\t\t/** Maximum latency in seconds at the 50th percentile. */\n\t\t\t\tp50?: number;\n\t\t\t\t/** Maximum latency in seconds at the 75th percentile. */\n\t\t\t\tp75?: number;\n\t\t\t\t/** Maximum latency in seconds at the 90th percentile. */\n\t\t\t\tp90?: number;\n\t\t\t\t/** Maximum latency in seconds at the 99th percentile. */\n\t\t\t\tp99?: number;\n\t\t  };\n}\n\n/**\n * Vercel AI Gateway routing preferences.\n * Controls which upstream providers the gateway routes requests to.\n * @see https://vercel.com/docs/ai-gateway/models-and-providers/provider-options\n */\nexport interface VercelGatewayRouting {\n\t/** List of provider slugs to exclusively use for this request (e.g., [\"bedrock\", \"anthropic\"]). */\n\tonly?: string[];\n\t/** List of provider slugs to try in order (e.g., [\"anthropic\", \"openai\"]). */\n\torder?: string[];\n}\n\n// Model interface for the unified model system\nexport interface Model<TApi extends Api> {\n\tid: string;\n\tname: string;\n\tapi: TApi;\n\tprovider: Provider;\n\tbaseUrl: string;\n\treasoning: boolean;\n\tinput: (\"text\" | \"image\")[];\n\tcost: {\n\t\tinput: number; // $/million tokens\n\t\toutput: number; // $/million tokens\n\t\tcacheRead: number; // $/million tokens\n\t\tcacheWrite: number; // $/million tokens\n\t};\n\tcontextWindow: number;\n\tmaxTokens: number;\n\theaders?: Record<string, string>;\n\t/** Compatibility overrides for OpenAI-compatible APIs. If not set, auto-detected from baseUrl. */\n\tcompat?: TApi extends \"openai-completions\"\n\t\t? OpenAICompletionsCompat\n\t\t: TApi extends \"openai-responses\"\n\t\t\t? OpenAIResponsesCompat\n\t\t\t: TApi extends \"anthropic-messages\"\n\t\t\t\t? AnthropicMessagesCompat\n\t\t\t\t: never;\n}\n"]}

package/dist/utils/oauth/index.d.ts

@@ -5,13 +5,9 @@
  * for OAuth-based providers:
  * - Anthropic (Claude Pro/Max)
  * - GitHub Copilot
- * - Google Cloud Code Assist (Gemini CLI)
- * - Antigravity (Gemini 3, Claude, GPT-OSS via Google Cloud)
  */
 export { anthropicOAuthProvider, loginAnthropic, refreshAnthropicToken } from "./anthropic.js";
 export { getGitHubCopilotBaseUrl, githubCopilotOAuthProvider, loginGitHubCopilot, normalizeDomain, refreshGitHubCopilotToken, } from "./github-copilot.js";
-export { antigravityOAuthProvider, loginAntigravity, refreshAntigravityToken } from "./google-antigravity.js";
-export { geminiCliOAuthProvider, loginGeminiCli, refreshGoogleCloudToken } from "./google-gemini-cli.js";
 export { loginOpenAICodex, openaiCodexOAuthProvider, refreshOpenAICodexToken } from "./openai-codex.js";
 export * from "./types.js";
 import type { OAuthCredentials, OAuthProviderId, OAuthProviderInfo, OAuthProviderInterface } from "./types.js";

package/dist/utils/oauth/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/utils/oauth/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAGH,OAAO,EAAE,sBAAsB,EAAE,cAAc,EAAE,qBAAqB,EAAE,MAAM,gBAAgB,CAAC;AAE/F,OAAO,EACN,uBAAuB,EACvB,0BAA0B,EAC1B,kBAAkB,EAClB,eAAe,EACf,yBAAyB,GACzB,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EAAE,wBAAwB,EAAE,gBAAgB,EAAE,uBAAuB,EAAE,MAAM,yBAAyB,CAAC;AAE9G,OAAO,EAAE,sBAAsB,EAAE,cAAc,EAAE,uBAAuB,EAAE,MAAM,wBAAwB,CAAC;AAEzG,OAAO,EAAE,gBAAgB,EAAE,wBAAwB,EAAE,uBAAuB,EAAE,MAAM,mBAAmB,CAAC;AAExG,cAAc,YAAY,CAAC;AAW3B,OAAO,KAAK,EAAE,gBAAgB,EAAE,eAAe,EAAE,iBAAiB,EAAE,sBAAsB,EAAE,MAAM,YAAY,CAAC;AAc/G;;GAEG;AACH,wBAAgB,gBAAgB,CAAC,EAAE,EAAE,eAAe,GAAG,sBAAsB,GAAG,SAAS,CAExF;AAED;;GAEG;AACH,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,sBAAsB,GAAG,IAAI,CAE5E;AAED;;;;;GAKG;AACH,wBAAgB,uBAAuB,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI,CAOxD;AAED;;GAEG;AACH,wBAAgB,mBAAmB,IAAI,IAAI,CAK1C;AAED;;GAEG;AACH,wBAAgB,iBAAiB,IAAI,sBAAsB,EAAE,CAE5D;AAED;;GAEG;AACH,wBAAgB,wBAAwB,IAAI,iBAAiB,EAAE,CAM9D;AAMD;;;GAGG;AACH,wBAAsB,iBAAiB,CACtC,UAAU,EAAE,eAAe,EAC3B,WAAW,EAAE,gBAAgB,GAC3B,OAAO,CAAC,gBAAgB,CAAC,CAM3B;AAED;;;;;;GAMG;AACH,wBAAsB,cAAc,CACnC,UAAU,EAAE,eAAe,EAC3B,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,gBAAgB,CAAC,GAC3C,OAAO,CAAC;IAAE,cAAc,EAAE,gBAAgB,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,GAAG,IAAI,CAAC,CAsBtE","sourcesContent":["/**\n * OAuth credential management for AI providers.\n *\n * This module handles login, token refresh, and credential storage\n * for OAuth-based providers:\n * - Anthropic (Claude Pro/Max)\n * - GitHub Copilot\n * - Google Cloud Code Assist (Gemini CLI)\n * - Antigravity (Gemini 3, Claude, GPT-OSS via Google Cloud)\n */\n\n// Anthropic\nexport { anthropicOAuthProvider, loginAnthropic, refreshAnthropicToken } from \"./anthropic.js\";\n// GitHub Copilot\nexport {\n\tgetGitHubCopilotBaseUrl,\n\tgithubCopilotOAuthProvider,\n\tloginGitHubCopilot,\n\tnormalizeDomain,\n\trefreshGitHubCopilotToken,\n} from \"./github-copilot.js\";\n// Google Antigravity\nexport { antigravityOAuthProvider, loginAntigravity, refreshAntigravityToken } from \"./google-antigravity.js\";\n// Google Gemini CLI\nexport { geminiCliOAuthProvider, loginGeminiCli, refreshGoogleCloudToken } from \"./google-gemini-cli.js\";\n// OpenAI Codex (ChatGPT OAuth)\nexport { loginOpenAICodex, openaiCodexOAuthProvider, refreshOpenAICodexToken } from \"./openai-codex.js\";\n\nexport * from \"./types.js\";\n\n// ============================================================================\n// Provider Registry\n// ============================================================================\n\nimport { anthropicOAuthProvider } from \"./anthropic.js\";\nimport { githubCopilotOAuthProvider } from \"./github-copilot.js\";\nimport { antigravityOAuthProvider } from \"./google-antigravity.js\";\nimport { geminiCliOAuthProvider } from \"./google-gemini-cli.js\";\nimport { openaiCodexOAuthProvider } from \"./openai-codex.js\";\nimport type { OAuthCredentials, OAuthProviderId, OAuthProviderInfo, OAuthProviderInterface } from \"./types.js\";\n\nconst BUILT_IN_OAUTH_PROVIDERS: OAuthProviderInterface[] = [\n\tanthropicOAuthProvider,\n\tgithubCopilotOAuthProvider,\n\tgeminiCliOAuthProvider,\n\tantigravityOAuthProvider,\n\topenaiCodexOAuthProvider,\n];\n\nconst oauthProviderRegistry = new Map<string, OAuthProviderInterface>(\n\tBUILT_IN_OAUTH_PROVIDERS.map((provider) => [provider.id, provider]),\n);\n\n/**\n * Get an OAuth provider by ID\n */\nexport function getOAuthProvider(id: OAuthProviderId): OAuthProviderInterface | undefined {\n\treturn oauthProviderRegistry.get(id);\n}\n\n/**\n * Register a custom OAuth provider\n */\nexport function registerOAuthProvider(provider: OAuthProviderInterface): void {\n\toauthProviderRegistry.set(provider.id, provider);\n}\n\n/**\n * Unregister an OAuth provider.\n *\n * If the provider is built-in, restores the built-in implementation.\n * Custom providers are removed completely.\n */\nexport function unregisterOAuthProvider(id: string): void {\n\tconst builtInProvider = BUILT_IN_OAUTH_PROVIDERS.find((provider) => provider.id === id);\n\tif (builtInProvider) {\n\t\toauthProviderRegistry.set(id, builtInProvider);\n\t\treturn;\n\t}\n\toauthProviderRegistry.delete(id);\n}\n\n/**\n * Reset OAuth providers to built-ins.\n */\nexport function resetOAuthProviders(): void {\n\toauthProviderRegistry.clear();\n\tfor (const provider of BUILT_IN_OAUTH_PROVIDERS) {\n\t\toauthProviderRegistry.set(provider.id, provider);\n\t}\n}\n\n/**\n * Get all registered OAuth providers\n */\nexport function getOAuthProviders(): OAuthProviderInterface[] {\n\treturn Array.from(oauthProviderRegistry.values());\n}\n\n/**\n * @deprecated Use getOAuthProviders() which returns OAuthProviderInterface[]\n */\nexport function getOAuthProviderInfoList(): OAuthProviderInfo[] {\n\treturn getOAuthProviders().map((p) => ({\n\t\tid: p.id,\n\t\tname: p.name,\n\t\tavailable: true,\n\t}));\n}\n\n// ============================================================================\n// High-level API (uses provider registry)\n// ============================================================================\n\n/**\n * Refresh token for any OAuth provider.\n * @deprecated Use getOAuthProvider(id).refreshToken() instead\n */\nexport async function refreshOAuthToken(\n\tproviderId: OAuthProviderId,\n\tcredentials: OAuthCredentials,\n): Promise<OAuthCredentials> {\n\tconst provider = getOAuthProvider(providerId);\n\tif (!provider) {\n\t\tthrow new Error(`Unknown OAuth provider: ${providerId}`);\n\t}\n\treturn provider.refreshToken(credentials);\n}\n\n/**\n * Get API key for a provider from OAuth credentials.\n * Automatically refreshes expired tokens.\n *\n * @returns API key string and updated credentials, or null if no credentials\n * @throws Error if refresh fails\n */\nexport async function getOAuthApiKey(\n\tproviderId: OAuthProviderId,\n\tcredentials: Record<string, OAuthCredentials>,\n): Promise<{ newCredentials: OAuthCredentials; apiKey: string } | null> {\n\tconst provider = getOAuthProvider(providerId);\n\tif (!provider) {\n\t\tthrow new Error(`Unknown OAuth provider: ${providerId}`);\n\t}\n\n\tlet creds = credentials[providerId];\n\tif (!creds) {\n\t\treturn null;\n\t}\n\n\t// Refresh if expired\n\tif (Date.now() >= creds.expires) {\n\t\ttry {\n\t\t\tcreds = await provider.refreshToken(creds);\n\t\t} catch (_error) {\n\t\t\tthrow new Error(`Failed to refresh OAuth token for ${providerId}`);\n\t\t}\n\t}\n\n\tconst apiKey = provider.getApiKey(creds);\n\treturn { newCredentials: creds, apiKey };\n}\n"]}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/utils/oauth/index.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAGH,OAAO,EAAE,sBAAsB,EAAE,cAAc,EAAE,qBAAqB,EAAE,MAAM,gBAAgB,CAAC;AAE/F,OAAO,EACN,uBAAuB,EACvB,0BAA0B,EAC1B,kBAAkB,EAClB,eAAe,EACf,yBAAyB,GACzB,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EAAE,gBAAgB,EAAE,wBAAwB,EAAE,uBAAuB,EAAE,MAAM,mBAAmB,CAAC;AAExG,cAAc,YAAY,CAAC;AAS3B,OAAO,KAAK,EAAE,gBAAgB,EAAE,eAAe,EAAE,iBAAiB,EAAE,sBAAsB,EAAE,MAAM,YAAY,CAAC;AAY/G;;GAEG;AACH,wBAAgB,gBAAgB,CAAC,EAAE,EAAE,eAAe,GAAG,sBAAsB,GAAG,SAAS,CAExF;AAED;;GAEG;AACH,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,sBAAsB,GAAG,IAAI,CAE5E;AAED;;;;;GAKG;AACH,wBAAgB,uBAAuB,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI,CAOxD;AAED;;GAEG;AACH,wBAAgB,mBAAmB,IAAI,IAAI,CAK1C;AAED;;GAEG;AACH,wBAAgB,iBAAiB,IAAI,sBAAsB,EAAE,CAE5D;AAED;;GAEG;AACH,wBAAgB,wBAAwB,IAAI,iBAAiB,EAAE,CAM9D;AAMD;;;GAGG;AACH,wBAAsB,iBAAiB,CACtC,UAAU,EAAE,eAAe,EAC3B,WAAW,EAAE,gBAAgB,GAC3B,OAAO,CAAC,gBAAgB,CAAC,CAM3B;AAED;;;;;;GAMG;AACH,wBAAsB,cAAc,CACnC,UAAU,EAAE,eAAe,EAC3B,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,gBAAgB,CAAC,GAC3C,OAAO,CAAC;IAAE,cAAc,EAAE,gBAAgB,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,GAAG,IAAI,CAAC,CAsBtE","sourcesContent":["/**\n * OAuth credential management for AI providers.\n *\n * This module handles login, token refresh, and credential storage\n * for OAuth-based providers:\n * - Anthropic (Claude Pro/Max)\n * - GitHub Copilot\n */\n\n// Anthropic\nexport { anthropicOAuthProvider, loginAnthropic, refreshAnthropicToken } from \"./anthropic.js\";\n// GitHub Copilot\nexport {\n\tgetGitHubCopilotBaseUrl,\n\tgithubCopilotOAuthProvider,\n\tloginGitHubCopilot,\n\tnormalizeDomain,\n\trefreshGitHubCopilotToken,\n} from \"./github-copilot.js\";\n// OpenAI Codex (ChatGPT OAuth)\nexport { loginOpenAICodex, openaiCodexOAuthProvider, refreshOpenAICodexToken } from \"./openai-codex.js\";\n\nexport * from \"./types.js\";\n\n// ============================================================================\n// Provider Registry\n// ============================================================================\n\nimport { anthropicOAuthProvider } from \"./anthropic.js\";\nimport { githubCopilotOAuthProvider } from \"./github-copilot.js\";\nimport { openaiCodexOAuthProvider } from \"./openai-codex.js\";\nimport type { OAuthCredentials, OAuthProviderId, OAuthProviderInfo, OAuthProviderInterface } from \"./types.js\";\n\nconst BUILT_IN_OAUTH_PROVIDERS: OAuthProviderInterface[] = [\n\tanthropicOAuthProvider,\n\tgithubCopilotOAuthProvider,\n\topenaiCodexOAuthProvider,\n];\n\nconst oauthProviderRegistry = new Map<string, OAuthProviderInterface>(\n\tBUILT_IN_OAUTH_PROVIDERS.map((provider) => [provider.id, provider]),\n);\n\n/**\n * Get an OAuth provider by ID\n */\nexport function getOAuthProvider(id: OAuthProviderId): OAuthProviderInterface | undefined {\n\treturn oauthProviderRegistry.get(id);\n}\n\n/**\n * Register a custom OAuth provider\n */\nexport function registerOAuthProvider(provider: OAuthProviderInterface): void {\n\toauthProviderRegistry.set(provider.id, provider);\n}\n\n/**\n * Unregister an OAuth provider.\n *\n * If the provider is built-in, restores the built-in implementation.\n * Custom providers are removed completely.\n */\nexport function unregisterOAuthProvider(id: string): void {\n\tconst builtInProvider = BUILT_IN_OAUTH_PROVIDERS.find((provider) => provider.id === id);\n\tif (builtInProvider) {\n\t\toauthProviderRegistry.set(id, builtInProvider);\n\t\treturn;\n\t}\n\toauthProviderRegistry.delete(id);\n}\n\n/**\n * Reset OAuth providers to built-ins.\n */\nexport function resetOAuthProviders(): void {\n\toauthProviderRegistry.clear();\n\tfor (const provider of BUILT_IN_OAUTH_PROVIDERS) {\n\t\toauthProviderRegistry.set(provider.id, provider);\n\t}\n}\n\n/**\n * Get all registered OAuth providers\n */\nexport function getOAuthProviders(): OAuthProviderInterface[] {\n\treturn Array.from(oauthProviderRegistry.values());\n}\n\n/**\n * @deprecated Use getOAuthProviders() which returns OAuthProviderInterface[]\n */\nexport function getOAuthProviderInfoList(): OAuthProviderInfo[] {\n\treturn getOAuthProviders().map((p) => ({\n\t\tid: p.id,\n\t\tname: p.name,\n\t\tavailable: true,\n\t}));\n}\n\n// ============================================================================\n// High-level API (uses provider registry)\n// ============================================================================\n\n/**\n * Refresh token for any OAuth provider.\n * @deprecated Use getOAuthProvider(id).refreshToken() instead\n */\nexport async function refreshOAuthToken(\n\tproviderId: OAuthProviderId,\n\tcredentials: OAuthCredentials,\n): Promise<OAuthCredentials> {\n\tconst provider = getOAuthProvider(providerId);\n\tif (!provider) {\n\t\tthrow new Error(`Unknown OAuth provider: ${providerId}`);\n\t}\n\treturn provider.refreshToken(credentials);\n}\n\n/**\n * Get API key for a provider from OAuth credentials.\n * Automatically refreshes expired tokens.\n *\n * @returns API key string and updated credentials, or null if no credentials\n * @throws Error if refresh fails\n */\nexport async function getOAuthApiKey(\n\tproviderId: OAuthProviderId,\n\tcredentials: Record<string, OAuthCredentials>,\n): Promise<{ newCredentials: OAuthCredentials; apiKey: string } | null> {\n\tconst provider = getOAuthProvider(providerId);\n\tif (!provider) {\n\t\tthrow new Error(`Unknown OAuth provider: ${providerId}`);\n\t}\n\n\tlet creds = credentials[providerId];\n\tif (!creds) {\n\t\treturn null;\n\t}\n\n\t// Refresh if expired\n\tif (Date.now() >= creds.expires) {\n\t\ttry {\n\t\t\tcreds = await provider.refreshToken(creds);\n\t\t} catch (_error) {\n\t\t\tthrow new Error(`Failed to refresh OAuth token for ${providerId}`);\n\t\t}\n\t}\n\n\tconst apiKey = provider.getApiKey(creds);\n\treturn { newCredentials: creds, apiKey };\n}\n"]}

package/dist/utils/oauth/index.js

@@ -5,17 +5,11 @@
  * for OAuth-based providers:
  * - Anthropic (Claude Pro/Max)
  * - GitHub Copilot
- * - Google Cloud Code Assist (Gemini CLI)
- * - Antigravity (Gemini 3, Claude, GPT-OSS via Google Cloud)
  */
 // Anthropic
 export { anthropicOAuthProvider, loginAnthropic, refreshAnthropicToken } from "./anthropic.js";
 // GitHub Copilot
 export { getGitHubCopilotBaseUrl, githubCopilotOAuthProvider, loginGitHubCopilot, normalizeDomain, refreshGitHubCopilotToken, } from "./github-copilot.js";
-// Google Antigravity
-export { antigravityOAuthProvider, loginAntigravity, refreshAntigravityToken } from "./google-antigravity.js";
-// Google Gemini CLI
-export { geminiCliOAuthProvider, loginGeminiCli, refreshGoogleCloudToken } from "./google-gemini-cli.js";
 // OpenAI Codex (ChatGPT OAuth)
 export { loginOpenAICodex, openaiCodexOAuthProvider, refreshOpenAICodexToken } from "./openai-codex.js";
 export * from "./types.js";
@@ -24,14 +18,10 @@ export * from "./types.js";
 // ============================================================================
 import { anthropicOAuthProvider } from "./anthropic.js";
 import { githubCopilotOAuthProvider } from "./github-copilot.js";
-import { antigravityOAuthProvider } from "./google-antigravity.js";
-import { geminiCliOAuthProvider } from "./google-gemini-cli.js";
 import { openaiCodexOAuthProvider } from "./openai-codex.js";
 const BUILT_IN_OAUTH_PROVIDERS = [
     anthropicOAuthProvider,
     githubCopilotOAuthProvider,
-    geminiCliOAuthProvider,
-    antigravityOAuthProvider,
     openaiCodexOAuthProvider,
 ];
 const oauthProviderRegistry = new Map(BUILT_IN_OAUTH_PROVIDERS.map((provider) => [provider.id, provider]));

package/dist/utils/oauth/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/utils/oauth/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,YAAY;AACZ,OAAO,EAAE,sBAAsB,EAAE,cAAc,EAAE,qBAAqB,EAAE,MAAM,gBAAgB,CAAC;AAC/F,iBAAiB;AACjB,OAAO,EACN,uBAAuB,EACvB,0BAA0B,EAC1B,kBAAkB,EAClB,eAAe,EACf,yBAAyB,GACzB,MAAM,qBAAqB,CAAC;AAC7B,qBAAqB;AACrB,OAAO,EAAE,wBAAwB,EAAE,gBAAgB,EAAE,uBAAuB,EAAE,MAAM,yBAAyB,CAAC;AAC9G,oBAAoB;AACpB,OAAO,EAAE,sBAAsB,EAAE,cAAc,EAAE,uBAAuB,EAAE,MAAM,wBAAwB,CAAC;AACzG,+BAA+B;AAC/B,OAAO,EAAE,gBAAgB,EAAE,wBAAwB,EAAE,uBAAuB,EAAE,MAAM,mBAAmB,CAAC;AAExG,cAAc,YAAY,CAAC;AAE3B,+EAA+E;AAC/E,oBAAoB;AACpB,+EAA+E;AAE/E,OAAO,EAAE,sBAAsB,EAAE,MAAM,gBAAgB,CAAC;AACxD,OAAO,EAAE,0BAA0B,EAAE,MAAM,qBAAqB,CAAC;AACjE,OAAO,EAAE,wBAAwB,EAAE,MAAM,yBAAyB,CAAC;AACnE,OAAO,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAC;AAChE,OAAO,EAAE,wBAAwB,EAAE,MAAM,mBAAmB,CAAC;AAG7D,MAAM,wBAAwB,GAA6B;IAC1D,sBAAsB;IACtB,0BAA0B;IAC1B,sBAAsB;IACtB,wBAAwB;IACxB,wBAAwB;CACxB,CAAC;AAEF,MAAM,qBAAqB,GAAG,IAAI,GAAG,CACpC,wBAAwB,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,EAAE,EAAE,QAAQ,CAAC,CAAC,CACnE,CAAC;AAEF;;GAEG;AACH,MAAM,UAAU,gBAAgB,CAAC,EAAmB,EAAsC;IACzF,OAAO,qBAAqB,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;AAAA,CACrC;AAED;;GAEG;AACH,MAAM,UAAU,qBAAqB,CAAC,QAAgC,EAAQ;IAC7E,qBAAqB,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,EAAE,QAAQ,CAAC,CAAC;AAAA,CACjD;AAED;;;;;GAKG;AACH,MAAM,UAAU,uBAAuB,CAAC,EAAU,EAAQ;IACzD,MAAM,eAAe,GAAG,wBAAwB,CAAC,IAAI,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC;IACxF,IAAI,eAAe,EAAE,CAAC;QACrB,qBAAqB,CAAC,GAAG,CAAC,EAAE,EAAE,eAAe,CAAC,CAAC;QAC/C,OAAO;IACR,CAAC;IACD,qBAAqB,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;AAAA,CACjC;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB,GAAS;IAC3C,qBAAqB,CAAC,KAAK,EAAE,CAAC;IAC9B,KAAK,MAAM,QAAQ,IAAI,wBAAwB,EAAE,CAAC;QACjD,qBAAqB,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,EAAE,QAAQ,CAAC,CAAC;IAClD,CAAC;AAAA,CACD;AAED;;GAEG;AACH,MAAM,UAAU,iBAAiB,GAA6B;IAC7D,OAAO,KAAK,CAAC,IAAI,CAAC,qBAAqB,CAAC,MAAM,EAAE,CAAC,CAAC;AAAA,CAClD;AAED;;GAEG;AACH,MAAM,UAAU,wBAAwB,GAAwB;IAC/D,OAAO,iBAAiB,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QACtC,EAAE,EAAE,CAAC,CAAC,EAAE;QACR,IAAI,EAAE,CAAC,CAAC,IAAI;QACZ,SAAS,EAAE,IAAI;KACf,CAAC,CAAC,CAAC;AAAA,CACJ;AAED,+EAA+E;AAC/E,0CAA0C;AAC1C,+EAA+E;AAE/E;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACtC,UAA2B,EAC3B,WAA6B,EACD;IAC5B,MAAM,QAAQ,GAAG,gBAAgB,CAAC,UAAU,CAAC,CAAC;IAC9C,IAAI,CAAC,QAAQ,EAAE,CAAC;QACf,MAAM,IAAI,KAAK,CAAC,2BAA2B,UAAU,EAAE,CAAC,CAAC;IAC1D,CAAC;IACD,OAAO,QAAQ,CAAC,YAAY,CAAC,WAAW,CAAC,CAAC;AAAA,CAC1C;AAED;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,cAAc,CACnC,UAA2B,EAC3B,WAA6C,EAC0B;IACvE,MAAM,QAAQ,GAAG,gBAAgB,CAAC,UAAU,CAAC,CAAC;IAC9C,IAAI,CAAC,QAAQ,EAAE,CAAC;QACf,MAAM,IAAI,KAAK,CAAC,2BAA2B,UAAU,EAAE,CAAC,CAAC;IAC1D,CAAC;IAED,IAAI,KAAK,GAAG,WAAW,CAAC,UAAU,CAAC,CAAC;IACpC,IAAI,CAAC,KAAK,EAAE,CAAC;QACZ,OAAO,IAAI,CAAC;IACb,CAAC;IAED,qBAAqB;IACrB,IAAI,IAAI,CAAC,GAAG,EAAE,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;QACjC,IAAI,CAAC;YACJ,KAAK,GAAG,MAAM,QAAQ,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC;QAC5C,CAAC;QAAC,OAAO,MAAM,EAAE,CAAC;YACjB,MAAM,IAAI,KAAK,CAAC,qCAAqC,UAAU,EAAE,CAAC,CAAC;QACpE,CAAC;IACF,CAAC;IAED,MAAM,MAAM,GAAG,QAAQ,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IACzC,OAAO,EAAE,cAAc,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC;AAAA,CACzC","sourcesContent":["/**\n * OAuth credential management for AI providers.\n *\n * This module handles login, token refresh, and credential storage\n * for OAuth-based providers:\n * - Anthropic (Claude Pro/Max)\n * - GitHub Copilot\n * - Google Cloud Code Assist (Gemini CLI)\n * - Antigravity (Gemini 3, Claude, GPT-OSS via Google Cloud)\n */\n\n// Anthropic\nexport { anthropicOAuthProvider, loginAnthropic, refreshAnthropicToken } from \"./anthropic.js\";\n// GitHub Copilot\nexport {\n\tgetGitHubCopilotBaseUrl,\n\tgithubCopilotOAuthProvider,\n\tloginGitHubCopilot,\n\tnormalizeDomain,\n\trefreshGitHubCopilotToken,\n} from \"./github-copilot.js\";\n// Google Antigravity\nexport { antigravityOAuthProvider, loginAntigravity, refreshAntigravityToken } from \"./google-antigravity.js\";\n// Google Gemini CLI\nexport { geminiCliOAuthProvider, loginGeminiCli, refreshGoogleCloudToken } from \"./google-gemini-cli.js\";\n// OpenAI Codex (ChatGPT OAuth)\nexport { loginOpenAICodex, openaiCodexOAuthProvider, refreshOpenAICodexToken } from \"./openai-codex.js\";\n\nexport * from \"./types.js\";\n\n// ============================================================================\n// Provider Registry\n// ============================================================================\n\nimport { anthropicOAuthProvider } from \"./anthropic.js\";\nimport { githubCopilotOAuthProvider } from \"./github-copilot.js\";\nimport { antigravityOAuthProvider } from \"./google-antigravity.js\";\nimport { geminiCliOAuthProvider } from \"./google-gemini-cli.js\";\nimport { openaiCodexOAuthProvider } from \"./openai-codex.js\";\nimport type { OAuthCredentials, OAuthProviderId, OAuthProviderInfo, OAuthProviderInterface } from \"./types.js\";\n\nconst BUILT_IN_OAUTH_PROVIDERS: OAuthProviderInterface[] = [\n\tanthropicOAuthProvider,\n\tgithubCopilotOAuthProvider,\n\tgeminiCliOAuthProvider,\n\tantigravityOAuthProvider,\n\topenaiCodexOAuthProvider,\n];\n\nconst oauthProviderRegistry = new Map<string, OAuthProviderInterface>(\n\tBUILT_IN_OAUTH_PROVIDERS.map((provider) => [provider.id, provider]),\n);\n\n/**\n * Get an OAuth provider by ID\n */\nexport function getOAuthProvider(id: OAuthProviderId): OAuthProviderInterface | undefined {\n\treturn oauthProviderRegistry.get(id);\n}\n\n/**\n * Register a custom OAuth provider\n */\nexport function registerOAuthProvider(provider: OAuthProviderInterface): void {\n\toauthProviderRegistry.set(provider.id, provider);\n}\n\n/**\n * Unregister an OAuth provider.\n *\n * If the provider is built-in, restores the built-in implementation.\n * Custom providers are removed completely.\n */\nexport function unregisterOAuthProvider(id: string): void {\n\tconst builtInProvider = BUILT_IN_OAUTH_PROVIDERS.find((provider) => provider.id === id);\n\tif (builtInProvider) {\n\t\toauthProviderRegistry.set(id, builtInProvider);\n\t\treturn;\n\t}\n\toauthProviderRegistry.delete(id);\n}\n\n/**\n * Reset OAuth providers to built-ins.\n */\nexport function resetOAuthProviders(): void {\n\toauthProviderRegistry.clear();\n\tfor (const provider of BUILT_IN_OAUTH_PROVIDERS) {\n\t\toauthProviderRegistry.set(provider.id, provider);\n\t}\n}\n\n/**\n * Get all registered OAuth providers\n */\nexport function getOAuthProviders(): OAuthProviderInterface[] {\n\treturn Array.from(oauthProviderRegistry.values());\n}\n\n/**\n * @deprecated Use getOAuthProviders() which returns OAuthProviderInterface[]\n */\nexport function getOAuthProviderInfoList(): OAuthProviderInfo[] {\n\treturn getOAuthProviders().map((p) => ({\n\t\tid: p.id,\n\t\tname: p.name,\n\t\tavailable: true,\n\t}));\n}\n\n// ============================================================================\n// High-level API (uses provider registry)\n// ============================================================================\n\n/**\n * Refresh token for any OAuth provider.\n * @deprecated Use getOAuthProvider(id).refreshToken() instead\n */\nexport async function refreshOAuthToken(\n\tproviderId: OAuthProviderId,\n\tcredentials: OAuthCredentials,\n): Promise<OAuthCredentials> {\n\tconst provider = getOAuthProvider(providerId);\n\tif (!provider) {\n\t\tthrow new Error(`Unknown OAuth provider: ${providerId}`);\n\t}\n\treturn provider.refreshToken(credentials);\n}\n\n/**\n * Get API key for a provider from OAuth credentials.\n * Automatically refreshes expired tokens.\n *\n * @returns API key string and updated credentials, or null if no credentials\n * @throws Error if refresh fails\n */\nexport async function getOAuthApiKey(\n\tproviderId: OAuthProviderId,\n\tcredentials: Record<string, OAuthCredentials>,\n): Promise<{ newCredentials: OAuthCredentials; apiKey: string } | null> {\n\tconst provider = getOAuthProvider(providerId);\n\tif (!provider) {\n\t\tthrow new Error(`Unknown OAuth provider: ${providerId}`);\n\t}\n\n\tlet creds = credentials[providerId];\n\tif (!creds) {\n\t\treturn null;\n\t}\n\n\t// Refresh if expired\n\tif (Date.now() >= creds.expires) {\n\t\ttry {\n\t\t\tcreds = await provider.refreshToken(creds);\n\t\t} catch (_error) {\n\t\t\tthrow new Error(`Failed to refresh OAuth token for ${providerId}`);\n\t\t}\n\t}\n\n\tconst apiKey = provider.getApiKey(creds);\n\treturn { newCredentials: creds, apiKey };\n}\n"]}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/utils/oauth/index.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,YAAY;AACZ,OAAO,EAAE,sBAAsB,EAAE,cAAc,EAAE,qBAAqB,EAAE,MAAM,gBAAgB,CAAC;AAC/F,iBAAiB;AACjB,OAAO,EACN,uBAAuB,EACvB,0BAA0B,EAC1B,kBAAkB,EAClB,eAAe,EACf,yBAAyB,GACzB,MAAM,qBAAqB,CAAC;AAC7B,+BAA+B;AAC/B,OAAO,EAAE,gBAAgB,EAAE,wBAAwB,EAAE,uBAAuB,EAAE,MAAM,mBAAmB,CAAC;AAExG,cAAc,YAAY,CAAC;AAE3B,+EAA+E;AAC/E,oBAAoB;AACpB,+EAA+E;AAE/E,OAAO,EAAE,sBAAsB,EAAE,MAAM,gBAAgB,CAAC;AACxD,OAAO,EAAE,0BAA0B,EAAE,MAAM,qBAAqB,CAAC;AACjE,OAAO,EAAE,wBAAwB,EAAE,MAAM,mBAAmB,CAAC;AAG7D,MAAM,wBAAwB,GAA6B;IAC1D,sBAAsB;IACtB,0BAA0B;IAC1B,wBAAwB;CACxB,CAAC;AAEF,MAAM,qBAAqB,GAAG,IAAI,GAAG,CACpC,wBAAwB,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,EAAE,EAAE,QAAQ,CAAC,CAAC,CACnE,CAAC;AAEF;;GAEG;AACH,MAAM,UAAU,gBAAgB,CAAC,EAAmB,EAAsC;IACzF,OAAO,qBAAqB,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;AAAA,CACrC;AAED;;GAEG;AACH,MAAM,UAAU,qBAAqB,CAAC,QAAgC,EAAQ;IAC7E,qBAAqB,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,EAAE,QAAQ,CAAC,CAAC;AAAA,CACjD;AAED;;;;;GAKG;AACH,MAAM,UAAU,uBAAuB,CAAC,EAAU,EAAQ;IACzD,MAAM,eAAe,GAAG,wBAAwB,CAAC,IAAI,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC;IACxF,IAAI,eAAe,EAAE,CAAC;QACrB,qBAAqB,CAAC,GAAG,CAAC,EAAE,EAAE,eAAe,CAAC,CAAC;QAC/C,OAAO;IACR,CAAC;IACD,qBAAqB,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;AAAA,CACjC;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB,GAAS;IAC3C,qBAAqB,CAAC,KAAK,EAAE,CAAC;IAC9B,KAAK,MAAM,QAAQ,IAAI,wBAAwB,EAAE,CAAC;QACjD,qBAAqB,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,EAAE,QAAQ,CAAC,CAAC;IAClD,CAAC;AAAA,CACD;AAED;;GAEG;AACH,MAAM,UAAU,iBAAiB,GAA6B;IAC7D,OAAO,KAAK,CAAC,IAAI,CAAC,qBAAqB,CAAC,MAAM,EAAE,CAAC,CAAC;AAAA,CAClD;AAED;;GAEG;AACH,MAAM,UAAU,wBAAwB,GAAwB;IAC/D,OAAO,iBAAiB,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QACtC,EAAE,EAAE,CAAC,CAAC,EAAE;QACR,IAAI,EAAE,CAAC,CAAC,IAAI;QACZ,SAAS,EAAE,IAAI;KACf,CAAC,CAAC,CAAC;AAAA,CACJ;AAED,+EAA+E;AAC/E,0CAA0C;AAC1C,+EAA+E;AAE/E;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACtC,UAA2B,EAC3B,WAA6B,EACD;IAC5B,MAAM,QAAQ,GAAG,gBAAgB,CAAC,UAAU,CAAC,CAAC;IAC9C,IAAI,CAAC,QAAQ,EAAE,CAAC;QACf,MAAM,IAAI,KAAK,CAAC,2BAA2B,UAAU,EAAE,CAAC,CAAC;IAC1D,CAAC;IACD,OAAO,QAAQ,CAAC,YAAY,CAAC,WAAW,CAAC,CAAC;AAAA,CAC1C;AAED;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,cAAc,CACnC,UAA2B,EAC3B,WAA6C,EAC0B;IACvE,MAAM,QAAQ,GAAG,gBAAgB,CAAC,UAAU,CAAC,CAAC;IAC9C,IAAI,CAAC,QAAQ,EAAE,CAAC;QACf,MAAM,IAAI,KAAK,CAAC,2BAA2B,UAAU,EAAE,CAAC,CAAC;IAC1D,CAAC;IAED,IAAI,KAAK,GAAG,WAAW,CAAC,UAAU,CAAC,CAAC;IACpC,IAAI,CAAC,KAAK,EAAE,CAAC;QACZ,OAAO,IAAI,CAAC;IACb,CAAC;IAED,qBAAqB;IACrB,IAAI,IAAI,CAAC,GAAG,EAAE,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;QACjC,IAAI,CAAC;YACJ,KAAK,GAAG,MAAM,QAAQ,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC;QAC5C,CAAC;QAAC,OAAO,MAAM,EAAE,CAAC;YACjB,MAAM,IAAI,KAAK,CAAC,qCAAqC,UAAU,EAAE,CAAC,CAAC;QACpE,CAAC;IACF,CAAC;IAED,MAAM,MAAM,GAAG,QAAQ,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IACzC,OAAO,EAAE,cAAc,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC;AAAA,CACzC","sourcesContent":["/**\n * OAuth credential management for AI providers.\n *\n * This module handles login, token refresh, and credential storage\n * for OAuth-based providers:\n * - Anthropic (Claude Pro/Max)\n * - GitHub Copilot\n */\n\n// Anthropic\nexport { anthropicOAuthProvider, loginAnthropic, refreshAnthropicToken } from \"./anthropic.js\";\n// GitHub Copilot\nexport {\n\tgetGitHubCopilotBaseUrl,\n\tgithubCopilotOAuthProvider,\n\tloginGitHubCopilot,\n\tnormalizeDomain,\n\trefreshGitHubCopilotToken,\n} from \"./github-copilot.js\";\n// OpenAI Codex (ChatGPT OAuth)\nexport { loginOpenAICodex, openaiCodexOAuthProvider, refreshOpenAICodexToken } from \"./openai-codex.js\";\n\nexport * from \"./types.js\";\n\n// ============================================================================\n// Provider Registry\n// ============================================================================\n\nimport { anthropicOAuthProvider } from \"./anthropic.js\";\nimport { githubCopilotOAuthProvider } from \"./github-copilot.js\";\nimport { openaiCodexOAuthProvider } from \"./openai-codex.js\";\nimport type { OAuthCredentials, OAuthProviderId, OAuthProviderInfo, OAuthProviderInterface } from \"./types.js\";\n\nconst BUILT_IN_OAUTH_PROVIDERS: OAuthProviderInterface[] = [\n\tanthropicOAuthProvider,\n\tgithubCopilotOAuthProvider,\n\topenaiCodexOAuthProvider,\n];\n\nconst oauthProviderRegistry = new Map<string, OAuthProviderInterface>(\n\tBUILT_IN_OAUTH_PROVIDERS.map((provider) => [provider.id, provider]),\n);\n\n/**\n * Get an OAuth provider by ID\n */\nexport function getOAuthProvider(id: OAuthProviderId): OAuthProviderInterface | undefined {\n\treturn oauthProviderRegistry.get(id);\n}\n\n/**\n * Register a custom OAuth provider\n */\nexport function registerOAuthProvider(provider: OAuthProviderInterface): void {\n\toauthProviderRegistry.set(provider.id, provider);\n}\n\n/**\n * Unregister an OAuth provider.\n *\n * If the provider is built-in, restores the built-in implementation.\n * Custom providers are removed completely.\n */\nexport function unregisterOAuthProvider(id: string): void {\n\tconst builtInProvider = BUILT_IN_OAUTH_PROVIDERS.find((provider) => provider.id === id);\n\tif (builtInProvider) {\n\t\toauthProviderRegistry.set(id, builtInProvider);\n\t\treturn;\n\t}\n\toauthProviderRegistry.delete(id);\n}\n\n/**\n * Reset OAuth providers to built-ins.\n */\nexport function resetOAuthProviders(): void {\n\toauthProviderRegistry.clear();\n\tfor (const provider of BUILT_IN_OAUTH_PROVIDERS) {\n\t\toauthProviderRegistry.set(provider.id, provider);\n\t}\n}\n\n/**\n * Get all registered OAuth providers\n */\nexport function getOAuthProviders(): OAuthProviderInterface[] {\n\treturn Array.from(oauthProviderRegistry.values());\n}\n\n/**\n * @deprecated Use getOAuthProviders() which returns OAuthProviderInterface[]\n */\nexport function getOAuthProviderInfoList(): OAuthProviderInfo[] {\n\treturn getOAuthProviders().map((p) => ({\n\t\tid: p.id,\n\t\tname: p.name,\n\t\tavailable: true,\n\t}));\n}\n\n// ============================================================================\n// High-level API (uses provider registry)\n// ============================================================================\n\n/**\n * Refresh token for any OAuth provider.\n * @deprecated Use getOAuthProvider(id).refreshToken() instead\n */\nexport async function refreshOAuthToken(\n\tproviderId: OAuthProviderId,\n\tcredentials: OAuthCredentials,\n): Promise<OAuthCredentials> {\n\tconst provider = getOAuthProvider(providerId);\n\tif (!provider) {\n\t\tthrow new Error(`Unknown OAuth provider: ${providerId}`);\n\t}\n\treturn provider.refreshToken(credentials);\n}\n\n/**\n * Get API key for a provider from OAuth credentials.\n * Automatically refreshes expired tokens.\n *\n * @returns API key string and updated credentials, or null if no credentials\n * @throws Error if refresh fails\n */\nexport async function getOAuthApiKey(\n\tproviderId: OAuthProviderId,\n\tcredentials: Record<string, OAuthCredentials>,\n): Promise<{ newCredentials: OAuthCredentials; apiKey: string } | null> {\n\tconst provider = getOAuthProvider(providerId);\n\tif (!provider) {\n\t\tthrow new Error(`Unknown OAuth provider: ${providerId}`);\n\t}\n\n\tlet creds = credentials[providerId];\n\tif (!creds) {\n\t\treturn null;\n\t}\n\n\t// Refresh if expired\n\tif (Date.now() >= creds.expires) {\n\t\ttry {\n\t\t\tcreds = await provider.refreshToken(creds);\n\t\t} catch (_error) {\n\t\t\tthrow new Error(`Failed to refresh OAuth token for ${providerId}`);\n\t\t}\n\t}\n\n\tconst apiKey = provider.getApiKey(creds);\n\treturn { newCredentials: creds, apiKey };\n}\n"]}

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@mariozechner/pi-ai",
-	"version": "0.70.6",
+	"version": "0.71.1",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"type": "module",
 	"main": "./dist/index.js",
@@ -22,10 +22,6 @@
 			"types": "./dist/providers/google.d.ts",
 			"import": "./dist/providers/google.js"
 		},
-		"./google-gemini-cli": {
-			"types": "./dist/providers/google-gemini-cli.d.ts",
-			"import": "./dist/providers/google-gemini-cli.js"
-		},
 		"./google-vertex": {
 			"types": "./dist/providers/google-vertex.d.ts",
 			"import": "./dist/providers/google-vertex.js"
@@ -72,7 +68,7 @@
 		"prepublishOnly": "npm run clean && npm run build"
 	},
 	"dependencies": {
-		"@anthropic-ai/sdk": "^0.90.0",
+		"@anthropic-ai/sdk": "^0.91.1",
 		"@aws-sdk/client-bedrock-runtime": "^3.1030.0",
 		"@google/genai": "^1.40.0",
 		"@mistralai/mistralai": "^2.2.0",

package/dist/providers/google-gemini-cli.d.ts

@@ -1,74 +0,0 @@
-/**
- * Google Gemini CLI / Antigravity provider.
- * Shared implementation for both google-gemini-cli and google-antigravity providers.
- * Uses the Cloud Code Assist API endpoint to access Gemini and Claude models.
- */
-import type { Content, ThinkingConfig } from "@google/genai";
-import type { Context, Model, SimpleStreamOptions, StreamFunction, StreamOptions } from "../types.js";
-import { convertTools, mapToolChoice } from "./google-shared.js";
-/**
- * Thinking level for Gemini 3 models.
- * Mirrors Google's ThinkingLevel enum values.
- */
-export type GoogleThinkingLevel = "THINKING_LEVEL_UNSPECIFIED" | "MINIMAL" | "LOW" | "MEDIUM" | "HIGH";
-export interface GoogleGeminiCliOptions extends StreamOptions {
-    toolChoice?: "auto" | "none" | "any";
-    /**
-     * Thinking/reasoning configuration.
-     * - Gemini 2.x models: use `budgetTokens` to set the thinking budget
-     * - Gemini 3 models (gemini-3-pro-*, gemini-3-flash-*): use `level` instead
-     *
-     * When using `streamSimple`, this is handled automatically based on the model.
-     */
-    thinking?: {
-        enabled: boolean;
-        /** Thinking budget in tokens. Use for Gemini 2.x models. */
-        budgetTokens?: number;
-        /** Thinking level. Use for Gemini 3 models (LOW/HIGH for Pro, MINIMAL/LOW/MEDIUM/HIGH for Flash). */
-        level?: GoogleThinkingLevel;
-    };
-    projectId?: string;
-}
-/**
- * Extract retry delay from Gemini error response (in milliseconds).
- * Checks headers first (Retry-After, x-ratelimit-reset, x-ratelimit-reset-after),
- * then parses body patterns like:
- * - "Your quota will reset after 39s"
- * - "Your quota will reset after 18h31m10s"
- * - "Please retry in Xs" or "Please retry in Xms"
- * - "retryDelay": "34.074824224s" (JSON field)
- */
-export declare function extractRetryDelay(errorText: string, response?: Response | Headers): number | undefined;
-interface CloudCodeAssistRequest {
-    project: string;
-    model: string;
-    request: {
-        contents: Content[];
-        sessionId?: string;
-        systemInstruction?: {
-            role?: string;
-            parts: {
-                text: string;
-            }[];
-        };
-        generationConfig?: {
-            maxOutputTokens?: number;
-            temperature?: number;
-            thinkingConfig?: ThinkingConfig;
-        };
-        tools?: ReturnType<typeof convertTools>;
-        toolConfig?: {
-            functionCallingConfig: {
-                mode: ReturnType<typeof mapToolChoice>;
-            };
-        };
-    };
-    requestType?: string;
-    userAgent?: string;
-    requestId?: string;
-}
-export declare const streamGoogleGeminiCli: StreamFunction<"google-gemini-cli", GoogleGeminiCliOptions>;
-export declare const streamSimpleGoogleGeminiCli: StreamFunction<"google-gemini-cli", SimpleStreamOptions>;
-export declare function buildRequest(model: Model<"google-gemini-cli">, context: Context, projectId: string, options?: GoogleGeminiCliOptions, isAntigravity?: boolean): CloudCodeAssistRequest;
-export {};
-//# sourceMappingURL=google-gemini-cli.d.ts.map

package/dist/providers/google-gemini-cli.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"google-gemini-cli.d.ts","sourceRoot":"","sources":["../../src/providers/google-gemini-cli.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,OAAO,EAAE,cAAc,EAAE,MAAM,eAAe,CAAC;AAE7D,OAAO,KAAK,EAGX,OAAO,EACP,KAAK,EACL,mBAAmB,EACnB,cAAc,EACd,aAAa,EAMb,MAAM,aAAa,CAAC;AAIrB,OAAO,EAEN,YAAY,EAGZ,aAAa,EAEb,MAAM,oBAAoB,CAAC;AAG5B;;;GAGG;AACH,MAAM,MAAM,mBAAmB,GAAG,4BAA4B,GAAG,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;AAEvG,MAAM,WAAW,sBAAuB,SAAQ,aAAa;IAC5D,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,KAAK,CAAC;IACrC;;;;;;OAMG;IACH,QAAQ,CAAC,EAAE;QACV,OAAO,EAAE,OAAO,CAAC;QACjB,4DAA4D;QAC5D,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,qGAAqG;QACrG,KAAK,CAAC,EAAE,mBAAmB,CAAC;KAC5B,CAAC;IACF,SAAS,CAAC,EAAE,MAAM,CAAC;CACnB;AAgDD;;;;;;;;GAQG;AACH,wBAAgB,iBAAiB,CAAC,SAAS,EAAE,MAAM,EAAE,QAAQ,CAAC,EAAE,QAAQ,GAAG,OAAO,GAAG,MAAM,GAAG,SAAS,CAyFtG;AA6DD,UAAU,sBAAsB;IAC/B,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE;QACR,QAAQ,EAAE,OAAO,EAAE,CAAC;QACpB,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,iBAAiB,CAAC,EAAE;YAAE,IAAI,CAAC,EAAE,MAAM,CAAC;YAAC,KAAK,EAAE;gBAAE,IAAI,EAAE,MAAM,CAAA;aAAE,EAAE,CAAA;SAAE,CAAC;QACjE,gBAAgB,CAAC,EAAE;YAClB,eAAe,CAAC,EAAE,MAAM,CAAC;YACzB,WAAW,CAAC,EAAE,MAAM,CAAC;YACrB,cAAc,CAAC,EAAE,cAAc,CAAC;SAChC,CAAC;QACF,KAAK,CAAC,EAAE,UAAU,CAAC,OAAO,YAAY,CAAC,CAAC;QACxC,UAAU,CAAC,EAAE;YACZ,qBAAqB,EAAE;gBACtB,IAAI,EAAE,UAAU,CAAC,OAAO,aAAa,CAAC,CAAC;aACvC,CAAC;SACF,CAAC;KACF,CAAC;IACF,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,SAAS,CAAC,EAAE,MAAM,CAAC;CACnB;AAiCD,eAAO,MAAM,qBAAqB,EAAE,cAAc,CAAC,mBAAmB,EAAE,sBAAsB,CA8e7F,CAAC;AAEF,eAAO,MAAM,2BAA2B,EAAE,cAAc,CAAC,mBAAmB,EAAE,mBAAmB,CAqDhG,CAAC;AAEF,wBAAgB,YAAY,CAC3B,KAAK,EAAE,KAAK,CAAC,mBAAmB,CAAC,EACjC,OAAO,EAAE,OAAO,EAChB,SAAS,EAAE,MAAM,EACjB,OAAO,GAAE,sBAA2B,EACpC,aAAa,UAAQ,GACnB,sBAAsB,CA8ExB","sourcesContent":["/**\n * Google Gemini CLI / Antigravity provider.\n * Shared implementation for both google-gemini-cli and google-antigravity providers.\n * Uses the Cloud Code Assist API endpoint to access Gemini and Claude models.\n */\n\nimport type { Content, ThinkingConfig } from \"@google/genai\";\nimport { calculateCost } from \"../models.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tContext,\n\tModel,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tTextContent,\n\tThinkingBudgets,\n\tThinkingContent,\n\tThinkingLevel,\n\tToolCall,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { headersToRecord } from \"../utils/headers.js\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.js\";\nimport {\n\tconvertMessages,\n\tconvertTools,\n\tisThinkingPart,\n\tmapStopReasonString,\n\tmapToolChoice,\n\tretainThoughtSignature,\n} from \"./google-shared.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\n\n/**\n * Thinking level for Gemini 3 models.\n * Mirrors Google's ThinkingLevel enum values.\n */\nexport type GoogleThinkingLevel = \"THINKING_LEVEL_UNSPECIFIED\" | \"MINIMAL\" | \"LOW\" | \"MEDIUM\" | \"HIGH\";\n\nexport interface GoogleGeminiCliOptions extends StreamOptions {\n\ttoolChoice?: \"auto\" | \"none\" | \"any\";\n\t/**\n\t * Thinking/reasoning configuration.\n\t * - Gemini 2.x models: use `budgetTokens` to set the thinking budget\n\t * - Gemini 3 models (gemini-3-pro-*, gemini-3-flash-*): use `level` instead\n\t *\n\t * When using `streamSimple`, this is handled automatically based on the model.\n\t */\n\tthinking?: {\n\t\tenabled: boolean;\n\t\t/** Thinking budget in tokens. Use for Gemini 2.x models. */\n\t\tbudgetTokens?: number;\n\t\t/** Thinking level. Use for Gemini 3 models (LOW/HIGH for Pro, MINIMAL/LOW/MEDIUM/HIGH for Flash). */\n\t\tlevel?: GoogleThinkingLevel;\n\t};\n\tprojectId?: string;\n}\n\nconst DEFAULT_ENDPOINT = \"https://cloudcode-pa.googleapis.com\";\nconst ANTIGRAVITY_DAILY_ENDPOINT = \"https://daily-cloudcode-pa.sandbox.googleapis.com\";\nconst ANTIGRAVITY_AUTOPUSH_ENDPOINT = \"https://autopush-cloudcode-pa.sandbox.googleapis.com\";\nconst ANTIGRAVITY_ENDPOINT_FALLBACKS = [\n\tANTIGRAVITY_DAILY_ENDPOINT,\n\tANTIGRAVITY_AUTOPUSH_ENDPOINT,\n\tDEFAULT_ENDPOINT,\n] as const;\n// Headers for Gemini CLI (prod endpoint)\nconst GEMINI_CLI_HEADERS = {\n\t\"User-Agent\": \"google-cloud-sdk vscode_cloudshelleditor/0.1\",\n\t\"X-Goog-Api-Client\": \"gl-node/22.17.0\",\n\t\"Client-Metadata\": JSON.stringify({\n\t\tideType: \"IDE_UNSPECIFIED\",\n\t\tplatform: \"PLATFORM_UNSPECIFIED\",\n\t\tpluginType: \"GEMINI\",\n\t}),\n};\n\n// Headers for Antigravity (sandbox endpoint) - requires specific User-Agent\nconst DEFAULT_ANTIGRAVITY_VERSION = \"1.21.9\";\n\nfunction getAntigravityHeaders() {\n\tconst version = process.env.PI_AI_ANTIGRAVITY_VERSION || DEFAULT_ANTIGRAVITY_VERSION;\n\treturn {\n\t\t\"User-Agent\": `antigravity/${version} darwin/arm64`,\n\t};\n}\n\n// Antigravity system instruction (compact version from CLIProxyAPI).\nconst ANTIGRAVITY_SYSTEM_INSTRUCTION =\n\t\"You are Antigravity, a powerful agentic AI coding assistant designed by the Google Deepmind team working on Advanced Agentic Coding.\" +\n\t\"You are pair programming with a USER to solve their coding task. The task may require creating a new codebase, modifying or debugging an existing codebase, or simply answering a question.\" +\n\t\"**Absolute paths only**\" +\n\t\"**Proactiveness**\";\n\n// Counter for generating unique tool call IDs\nlet toolCallCounter = 0;\n\n// Retry configuration\nconst MAX_RETRIES = 3;\nconst BASE_DELAY_MS = 1000;\nconst MAX_EMPTY_STREAM_RETRIES = 2;\nconst EMPTY_STREAM_BASE_DELAY_MS = 500;\nconst CLAUDE_THINKING_BETA_HEADER = \"interleaved-thinking-2025-05-14\";\n\n/**\n * Extract retry delay from Gemini error response (in milliseconds).\n * Checks headers first (Retry-After, x-ratelimit-reset, x-ratelimit-reset-after),\n * then parses body patterns like:\n * - \"Your quota will reset after 39s\"\n * - \"Your quota will reset after 18h31m10s\"\n * - \"Please retry in Xs\" or \"Please retry in Xms\"\n * - \"retryDelay\": \"34.074824224s\" (JSON field)\n */\nexport function extractRetryDelay(errorText: string, response?: Response | Headers): number | undefined {\n\tconst normalizeDelay = (ms: number): number | undefined => (ms > 0 ? Math.ceil(ms + 1000) : undefined);\n\n\tconst headers = response instanceof Headers ? response : response?.headers;\n\tif (headers) {\n\t\tconst retryAfter = headers.get(\"retry-after\");\n\t\tif (retryAfter) {\n\t\t\tconst retryAfterSeconds = Number(retryAfter);\n\t\t\tif (Number.isFinite(retryAfterSeconds)) {\n\t\t\t\tconst delay = normalizeDelay(retryAfterSeconds * 1000);\n\t\t\t\tif (delay !== undefined) {\n\t\t\t\t\treturn delay;\n\t\t\t\t}\n\t\t\t}\n\t\t\tconst retryAfterDate = new Date(retryAfter);\n\t\t\tconst retryAfterMs = retryAfterDate.getTime();\n\t\t\tif (!Number.isNaN(retryAfterMs)) {\n\t\t\t\tconst delay = normalizeDelay(retryAfterMs - Date.now());\n\t\t\t\tif (delay !== undefined) {\n\t\t\t\t\treturn delay;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tconst rateLimitReset = headers.get(\"x-ratelimit-reset\");\n\t\tif (rateLimitReset) {\n\t\t\tconst resetSeconds = Number.parseInt(rateLimitReset, 10);\n\t\t\tif (!Number.isNaN(resetSeconds)) {\n\t\t\t\tconst delay = normalizeDelay(resetSeconds * 1000 - Date.now());\n\t\t\t\tif (delay !== undefined) {\n\t\t\t\t\treturn delay;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tconst rateLimitResetAfter = headers.get(\"x-ratelimit-reset-after\");\n\t\tif (rateLimitResetAfter) {\n\t\t\tconst resetAfterSeconds = Number(rateLimitResetAfter);\n\t\t\tif (Number.isFinite(resetAfterSeconds)) {\n\t\t\t\tconst delay = normalizeDelay(resetAfterSeconds * 1000);\n\t\t\t\tif (delay !== undefined) {\n\t\t\t\t\treturn delay;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Pattern 1: \"Your quota will reset after ...\" (formats: \"18h31m10s\", \"10m15s\", \"6s\", \"39s\")\n\tconst durationMatch = errorText.match(/reset after (?:(\\d+)h)?(?:(\\d+)m)?(\\d+(?:\\.\\d+)?)s/i);\n\tif (durationMatch) {\n\t\tconst hours = durationMatch[1] ? parseInt(durationMatch[1], 10) : 0;\n\t\tconst minutes = durationMatch[2] ? parseInt(durationMatch[2], 10) : 0;\n\t\tconst seconds = parseFloat(durationMatch[3]);\n\t\tif (!Number.isNaN(seconds)) {\n\t\t\tconst totalMs = ((hours * 60 + minutes) * 60 + seconds) * 1000;\n\t\t\tconst delay = normalizeDelay(totalMs);\n\t\t\tif (delay !== undefined) {\n\t\t\t\treturn delay;\n\t\t\t}\n\t\t}\n\t}\n\n\t// Pattern 2: \"Please retry in X[ms|s]\"\n\tconst retryInMatch = errorText.match(/Please retry in ([0-9.]+)(ms|s)/i);\n\tif (retryInMatch?.[1]) {\n\t\tconst value = parseFloat(retryInMatch[1]);\n\t\tif (!Number.isNaN(value) && value > 0) {\n\t\t\tconst ms = retryInMatch[2].toLowerCase() === \"ms\" ? value : value * 1000;\n\t\t\tconst delay = normalizeDelay(ms);\n\t\t\tif (delay !== undefined) {\n\t\t\t\treturn delay;\n\t\t\t}\n\t\t}\n\t}\n\n\t// Pattern 3: \"retryDelay\": \"34.074824224s\" (JSON field in error details)\n\tconst retryDelayMatch = errorText.match(/\"retryDelay\":\\s*\"([0-9.]+)(ms|s)\"/i);\n\tif (retryDelayMatch?.[1]) {\n\t\tconst value = parseFloat(retryDelayMatch[1]);\n\t\tif (!Number.isNaN(value) && value > 0) {\n\t\t\tconst ms = retryDelayMatch[2].toLowerCase() === \"ms\" ? value : value * 1000;\n\t\t\tconst delay = normalizeDelay(ms);\n\t\t\tif (delay !== undefined) {\n\t\t\t\treturn delay;\n\t\t\t}\n\t\t}\n\t}\n\n\treturn undefined;\n}\n\nfunction needsClaudeThinkingBetaHeader(model: Model<\"google-gemini-cli\">): boolean {\n\treturn model.provider === \"google-antigravity\" && model.id.startsWith(\"claude-\") && model.reasoning;\n}\n\nfunction isGemini3ProModel(modelId: string): boolean {\n\treturn /gemini-3(?:\\.1)?-pro/.test(modelId.toLowerCase());\n}\n\nfunction isGemini3FlashModel(modelId: string): boolean {\n\treturn /gemini-3(?:\\.1)?-flash/.test(modelId.toLowerCase());\n}\n\nfunction isGemini3Model(modelId: string): boolean {\n\treturn isGemini3ProModel(modelId) || isGemini3FlashModel(modelId);\n}\n\n/**\n * Check if an error is retryable (rate limit, server error, network error, etc.)\n */\nfunction isRetryableError(status: number, errorText: string): boolean {\n\tif (status === 429 || status === 500 || status === 502 || status === 503 || status === 504) {\n\t\treturn true;\n\t}\n\treturn /resource.?exhausted|rate.?limit|overloaded|service.?unavailable|other.?side.?closed/i.test(errorText);\n}\n\n/**\n * Extract a clean, user-friendly error message from Google API error response.\n * Parses JSON error responses and returns just the message field.\n */\nfunction extractErrorMessage(errorText: string): string {\n\ttry {\n\t\tconst parsed = JSON.parse(errorText) as { error?: { message?: string } };\n\t\tif (parsed.error?.message) {\n\t\t\treturn parsed.error.message;\n\t\t}\n\t} catch {\n\t\t// Not JSON, return as-is\n\t}\n\treturn errorText;\n}\n\n/**\n * Sleep for a given number of milliseconds, respecting abort signal.\n */\nfunction sleep(ms: number, signal?: AbortSignal): Promise<void> {\n\treturn new Promise((resolve, reject) => {\n\t\tif (signal?.aborted) {\n\t\t\treject(new Error(\"Request was aborted\"));\n\t\t\treturn;\n\t\t}\n\t\tconst timeout = setTimeout(resolve, ms);\n\t\tsignal?.addEventListener(\"abort\", () => {\n\t\t\tclearTimeout(timeout);\n\t\t\treject(new Error(\"Request was aborted\"));\n\t\t});\n\t});\n}\n\ninterface CloudCodeAssistRequest {\n\tproject: string;\n\tmodel: string;\n\trequest: {\n\t\tcontents: Content[];\n\t\tsessionId?: string;\n\t\tsystemInstruction?: { role?: string; parts: { text: string }[] };\n\t\tgenerationConfig?: {\n\t\t\tmaxOutputTokens?: number;\n\t\t\ttemperature?: number;\n\t\t\tthinkingConfig?: ThinkingConfig;\n\t\t};\n\t\ttools?: ReturnType<typeof convertTools>;\n\t\ttoolConfig?: {\n\t\t\tfunctionCallingConfig: {\n\t\t\t\tmode: ReturnType<typeof mapToolChoice>;\n\t\t\t};\n\t\t};\n\t};\n\trequestType?: string;\n\tuserAgent?: string;\n\trequestId?: string;\n}\n\ninterface CloudCodeAssistResponseChunk {\n\tresponse?: {\n\t\tcandidates?: Array<{\n\t\t\tcontent?: {\n\t\t\t\trole: string;\n\t\t\t\tparts?: Array<{\n\t\t\t\t\ttext?: string;\n\t\t\t\t\tthought?: boolean;\n\t\t\t\t\tthoughtSignature?: string;\n\t\t\t\t\tfunctionCall?: {\n\t\t\t\t\t\tname: string;\n\t\t\t\t\t\targs: Record<string, unknown>;\n\t\t\t\t\t\tid?: string;\n\t\t\t\t\t};\n\t\t\t\t}>;\n\t\t\t};\n\t\t\tfinishReason?: string;\n\t\t}>;\n\t\tusageMetadata?: {\n\t\t\tpromptTokenCount?: number;\n\t\t\tcandidatesTokenCount?: number;\n\t\t\tthoughtsTokenCount?: number;\n\t\t\ttotalTokenCount?: number;\n\t\t\tcachedContentTokenCount?: number;\n\t\t};\n\t\tmodelVersion?: string;\n\t\tresponseId?: string;\n\t};\n\ttraceId?: string;\n}\n\nexport const streamGoogleGeminiCli: StreamFunction<\"google-gemini-cli\", GoogleGeminiCliOptions> = (\n\tmodel: Model<\"google-gemini-cli\">,\n\tcontext: Context,\n\toptions?: GoogleGeminiCliOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t(async () => {\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: \"google-gemini-cli\" as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\t// apiKey is JSON-encoded: { token, projectId }\n\t\t\tconst apiKeyRaw = options?.apiKey;\n\t\t\tif (!apiKeyRaw) {\n\t\t\t\tthrow new Error(\"Google Cloud Code Assist requires OAuth authentication. Use /login to authenticate.\");\n\t\t\t}\n\n\t\t\tlet accessToken: string;\n\t\t\tlet projectId: string;\n\n\t\t\ttry {\n\t\t\t\tconst parsed = JSON.parse(apiKeyRaw) as { token: string; projectId: string };\n\t\t\t\taccessToken = parsed.token;\n\t\t\t\tprojectId = parsed.projectId;\n\t\t\t} catch {\n\t\t\t\tthrow new Error(\"Invalid Google Cloud Code Assist credentials. Use /login to re-authenticate.\");\n\t\t\t}\n\n\t\t\tif (!accessToken || !projectId) {\n\t\t\t\tthrow new Error(\"Missing token or projectId in Google Cloud credentials. Use /login to re-authenticate.\");\n\t\t\t}\n\n\t\t\tconst isAntigravity = model.provider === \"google-antigravity\";\n\t\t\tconst baseUrl = model.baseUrl?.trim();\n\t\t\tconst endpoints = baseUrl ? [baseUrl] : isAntigravity ? ANTIGRAVITY_ENDPOINT_FALLBACKS : [DEFAULT_ENDPOINT];\n\n\t\t\tlet requestBody = buildRequest(model, context, projectId, options, isAntigravity);\n\t\t\tconst nextRequestBody = await options?.onPayload?.(requestBody, model);\n\t\t\tif (nextRequestBody !== undefined) {\n\t\t\t\trequestBody = nextRequestBody as CloudCodeAssistRequest;\n\t\t\t}\n\t\t\tconst headers = isAntigravity ? getAntigravityHeaders() : GEMINI_CLI_HEADERS;\n\n\t\t\tconst requestHeaders = {\n\t\t\t\tAuthorization: `Bearer ${accessToken}`,\n\t\t\t\t\"Content-Type\": \"application/json\",\n\t\t\t\tAccept: \"text/event-stream\",\n\t\t\t\t...headers,\n\t\t\t\t...(needsClaudeThinkingBetaHeader(model) ? { \"anthropic-beta\": CLAUDE_THINKING_BETA_HEADER } : {}),\n\t\t\t\t...options?.headers,\n\t\t\t};\n\t\t\tconst requestBodyJson = JSON.stringify(requestBody);\n\n\t\t\t// Fetch with retry logic for rate limits, transient errors, and endpoint fallbacks.\n\t\t\t// On 403/404, immediately try the next endpoint (no delay).\n\t\t\t// On 429/5xx, retry with backoff on the same or next endpoint.\n\t\t\tlet response: Response | undefined;\n\t\t\tlet lastError: Error | undefined;\n\t\t\tlet requestUrl: string | undefined;\n\t\t\tlet endpointIndex = 0;\n\n\t\t\tfor (let attempt = 0; attempt <= MAX_RETRIES; attempt++) {\n\t\t\t\tif (options?.signal?.aborted) {\n\t\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t\t}\n\n\t\t\t\ttry {\n\t\t\t\t\tconst endpoint = endpoints[endpointIndex];\n\t\t\t\t\trequestUrl = `${endpoint}/v1internal:streamGenerateContent?alt=sse`;\n\t\t\t\t\tresponse = await fetch(requestUrl, {\n\t\t\t\t\t\tmethod: \"POST\",\n\t\t\t\t\t\theaders: requestHeaders,\n\t\t\t\t\t\tbody: requestBodyJson,\n\t\t\t\t\t\tsignal: options?.signal,\n\t\t\t\t\t});\n\t\t\t\t\tawait options?.onResponse?.(\n\t\t\t\t\t\t{ status: response.status, headers: headersToRecord(response.headers) },\n\t\t\t\t\t\tmodel,\n\t\t\t\t\t);\n\n\t\t\t\t\tif (response.ok) {\n\t\t\t\t\t\tbreak; // Success, exit retry loop\n\t\t\t\t\t}\n\n\t\t\t\t\tconst errorText = await response.text();\n\n\t\t\t\t\t// On 403/404, cascade to the next endpoint immediately (no delay)\n\t\t\t\t\tif ((response.status === 403 || response.status === 404) && endpointIndex < endpoints.length - 1) {\n\t\t\t\t\t\tendpointIndex++;\n\t\t\t\t\t\tcontinue;\n\t\t\t\t\t}\n\n\t\t\t\t\t// Check if retryable (429, 5xx, network patterns)\n\t\t\t\t\tif (attempt < MAX_RETRIES && isRetryableError(response.status, errorText)) {\n\t\t\t\t\t\t// Advance endpoint if possible\n\t\t\t\t\t\tif (endpointIndex < endpoints.length - 1) {\n\t\t\t\t\t\t\tendpointIndex++;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// Use server-provided delay or exponential backoff\n\t\t\t\t\t\tconst serverDelay = extractRetryDelay(errorText, response);\n\t\t\t\t\t\tconst delayMs = serverDelay ?? BASE_DELAY_MS * 2 ** attempt;\n\n\t\t\t\t\t\t// Check if server delay exceeds max allowed (default: 60s)\n\t\t\t\t\t\tconst maxDelayMs = options?.maxRetryDelayMs ?? 60000;\n\t\t\t\t\t\tif (maxDelayMs > 0 && serverDelay && serverDelay > maxDelayMs) {\n\t\t\t\t\t\t\tconst delaySeconds = Math.ceil(serverDelay / 1000);\n\t\t\t\t\t\t\tthrow new Error(\n\t\t\t\t\t\t\t\t`Server requested ${delaySeconds}s retry delay (max: ${Math.ceil(maxDelayMs / 1000)}s). ${extractErrorMessage(errorText)}`,\n\t\t\t\t\t\t\t);\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\tawait sleep(delayMs, options?.signal);\n\t\t\t\t\t\tcontinue;\n\t\t\t\t\t}\n\n\t\t\t\t\t// Not retryable or max retries exceeded\n\t\t\t\t\tthrow new Error(`Cloud Code Assist API error (${response.status}): ${extractErrorMessage(errorText)}`);\n\t\t\t\t} catch (error) {\n\t\t\t\t\t// Check for abort - fetch throws AbortError, our code throws \"Request was aborted\"\n\t\t\t\t\tif (error instanceof Error) {\n\t\t\t\t\t\tif (error.name === \"AbortError\" || error.message === \"Request was aborted\") {\n\t\t\t\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\t// Extract detailed error message from fetch errors (Node includes cause)\n\t\t\t\t\tlastError = error instanceof Error ? error : new Error(String(error));\n\t\t\t\t\tif (lastError.message === \"fetch failed\" && lastError.cause instanceof Error) {\n\t\t\t\t\t\tlastError = new Error(`Network error: ${lastError.cause.message}`);\n\t\t\t\t\t}\n\t\t\t\t\t// Network errors are retryable\n\t\t\t\t\tif (attempt < MAX_RETRIES) {\n\t\t\t\t\t\tconst delayMs = BASE_DELAY_MS * 2 ** attempt;\n\t\t\t\t\t\tawait sleep(delayMs, options?.signal);\n\t\t\t\t\t\tcontinue;\n\t\t\t\t\t}\n\t\t\t\t\tthrow lastError;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (!response || !response.ok) {\n\t\t\t\tthrow lastError ?? new Error(\"Failed to get response after retries\");\n\t\t\t}\n\n\t\t\tlet started = false;\n\t\t\tconst ensureStarted = () => {\n\t\t\t\tif (!started) {\n\t\t\t\t\tstream.push({ type: \"start\", partial: output });\n\t\t\t\t\tstarted = true;\n\t\t\t\t}\n\t\t\t};\n\n\t\t\tconst resetOutput = () => {\n\t\t\t\toutput.content = [];\n\t\t\t\toutput.usage = {\n\t\t\t\t\tinput: 0,\n\t\t\t\t\toutput: 0,\n\t\t\t\t\tcacheRead: 0,\n\t\t\t\t\tcacheWrite: 0,\n\t\t\t\t\ttotalTokens: 0,\n\t\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t\t};\n\t\t\t\toutput.stopReason = \"stop\";\n\t\t\t\toutput.errorMessage = undefined;\n\t\t\t\toutput.timestamp = Date.now();\n\t\t\t\tstarted = false;\n\t\t\t};\n\n\t\t\tconst streamResponse = async (activeResponse: Response): Promise<boolean> => {\n\t\t\t\tif (!activeResponse.body) {\n\t\t\t\t\tthrow new Error(\"No response body\");\n\t\t\t\t}\n\n\t\t\t\tlet hasContent = false;\n\t\t\t\tlet currentBlock: TextContent | ThinkingContent | null = null;\n\t\t\t\tconst blocks = output.content;\n\t\t\t\tconst blockIndex = () => blocks.length - 1;\n\n\t\t\t\t// Read SSE stream\n\t\t\t\tconst reader = activeResponse.body.getReader();\n\t\t\t\tconst decoder = new TextDecoder();\n\t\t\t\tlet buffer = \"\";\n\n\t\t\t\t// Set up abort handler to cancel reader when signal fires\n\t\t\t\tconst abortHandler = () => {\n\t\t\t\t\tvoid reader.cancel().catch(() => {});\n\t\t\t\t};\n\t\t\t\toptions?.signal?.addEventListener(\"abort\", abortHandler);\n\n\t\t\t\ttry {\n\t\t\t\t\twhile (true) {\n\t\t\t\t\t\t// Check abort signal before each read\n\t\t\t\t\t\tif (options?.signal?.aborted) {\n\t\t\t\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\tconst { done, value } = await reader.read();\n\t\t\t\t\t\tif (done) break;\n\n\t\t\t\t\t\tbuffer += decoder.decode(value, { stream: true });\n\t\t\t\t\t\tconst lines = buffer.split(\"\\n\");\n\t\t\t\t\t\tbuffer = lines.pop() || \"\";\n\n\t\t\t\t\t\tfor (const line of lines) {\n\t\t\t\t\t\t\tif (!line.startsWith(\"data:\")) continue;\n\n\t\t\t\t\t\t\tconst jsonStr = line.slice(5).trim();\n\t\t\t\t\t\t\tif (!jsonStr) continue;\n\n\t\t\t\t\t\t\tlet chunk: CloudCodeAssistResponseChunk;\n\t\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\t\tchunk = JSON.parse(jsonStr);\n\t\t\t\t\t\t\t} catch {\n\t\t\t\t\t\t\t\tcontinue;\n\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\t// Unwrap the response\n\t\t\t\t\t\t\tconst responseData = chunk.response;\n\t\t\t\t\t\t\tif (!responseData) continue;\n\t\t\t\t\t\t\t// Cloud Code Assist mirrors Gemini's responseId field. Keep the first non-empty one.\n\t\t\t\t\t\t\t// A single streamed response should retain the same ID across chunks.\n\t\t\t\t\t\t\toutput.responseId ||= responseData.responseId;\n\n\t\t\t\t\t\t\tconst candidate = responseData.candidates?.[0];\n\t\t\t\t\t\t\tif (candidate?.content?.parts) {\n\t\t\t\t\t\t\t\tfor (const part of candidate.content.parts) {\n\t\t\t\t\t\t\t\t\tif (part.text !== undefined) {\n\t\t\t\t\t\t\t\t\t\thasContent = true;\n\t\t\t\t\t\t\t\t\t\tconst isThinking = isThinkingPart(part);\n\t\t\t\t\t\t\t\t\t\tif (\n\t\t\t\t\t\t\t\t\t\t\t!currentBlock ||\n\t\t\t\t\t\t\t\t\t\t\t(isThinking && currentBlock.type !== \"thinking\") ||\n\t\t\t\t\t\t\t\t\t\t\t(!isThinking && currentBlock.type !== \"text\")\n\t\t\t\t\t\t\t\t\t\t) {\n\t\t\t\t\t\t\t\t\t\t\tif (currentBlock) {\n\t\t\t\t\t\t\t\t\t\t\t\tif (currentBlock.type === \"text\") {\n\t\t\t\t\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\t\t\t\t\t\t\t\t\t\tcontentIndex: blocks.length - 1,\n\t\t\t\t\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.text,\n\t\t\t\t\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.thinking,\n\t\t\t\t\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t\t\tif (isThinking) {\n\t\t\t\t\t\t\t\t\t\t\t\tcurrentBlock = { type: \"thinking\", thinking: \"\", thinkingSignature: undefined };\n\t\t\t\t\t\t\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\t\t\t\t\t\t\tensureStarted();\n\t\t\t\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\t\t\t\ttype: \"thinking_start\",\n\t\t\t\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\t\t\t\tcurrentBlock = { type: \"text\", text: \"\" };\n\t\t\t\t\t\t\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\t\t\t\t\t\t\tensureStarted();\n\t\t\t\t\t\t\t\t\t\t\t\tstream.push({ type: \"text_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t\tif (currentBlock.type === \"thinking\") {\n\t\t\t\t\t\t\t\t\t\t\tcurrentBlock.thinking += part.text;\n\t\t\t\t\t\t\t\t\t\t\tcurrentBlock.thinkingSignature = retainThoughtSignature(\n\t\t\t\t\t\t\t\t\t\t\t\tcurrentBlock.thinkingSignature,\n\t\t\t\t\t\t\t\t\t\t\t\tpart.thoughtSignature,\n\t\t\t\t\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\t\t\ttype: \"thinking_delta\",\n\t\t\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\t\t\tdelta: part.text,\n\t\t\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\t\t\tcurrentBlock.text += part.text;\n\t\t\t\t\t\t\t\t\t\t\tcurrentBlock.textSignature = retainThoughtSignature(\n\t\t\t\t\t\t\t\t\t\t\t\tcurrentBlock.textSignature,\n\t\t\t\t\t\t\t\t\t\t\t\tpart.thoughtSignature,\n\t\t\t\t\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\t\t\ttype: \"text_delta\",\n\t\t\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\t\t\tdelta: part.text,\n\t\t\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\t\t\tif (part.functionCall) {\n\t\t\t\t\t\t\t\t\t\thasContent = true;\n\t\t\t\t\t\t\t\t\t\tif (currentBlock) {\n\t\t\t\t\t\t\t\t\t\t\tif (currentBlock.type === \"text\") {\n\t\t\t\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.text,\n\t\t\t\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.thinking,\n\t\t\t\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t\t\tcurrentBlock = null;\n\t\t\t\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\t\t\t\tconst providedId = part.functionCall.id;\n\t\t\t\t\t\t\t\t\t\tconst needsNewId =\n\t\t\t\t\t\t\t\t\t\t\t!providedId ||\n\t\t\t\t\t\t\t\t\t\t\toutput.content.some((b) => b.type === \"toolCall\" && b.id === providedId);\n\t\t\t\t\t\t\t\t\t\tconst toolCallId = needsNewId\n\t\t\t\t\t\t\t\t\t\t\t? `${part.functionCall.name}_${Date.now()}_${++toolCallCounter}`\n\t\t\t\t\t\t\t\t\t\t\t: providedId;\n\n\t\t\t\t\t\t\t\t\t\tconst toolCall: ToolCall = {\n\t\t\t\t\t\t\t\t\t\t\ttype: \"toolCall\",\n\t\t\t\t\t\t\t\t\t\t\tid: toolCallId,\n\t\t\t\t\t\t\t\t\t\t\tname: part.functionCall.name || \"\",\n\t\t\t\t\t\t\t\t\t\t\targuments: (part.functionCall.args as Record<string, unknown>) ?? {},\n\t\t\t\t\t\t\t\t\t\t\t...(part.thoughtSignature && { thoughtSignature: part.thoughtSignature }),\n\t\t\t\t\t\t\t\t\t\t};\n\n\t\t\t\t\t\t\t\t\t\toutput.content.push(toolCall);\n\t\t\t\t\t\t\t\t\t\tensureStarted();\n\t\t\t\t\t\t\t\t\t\tstream.push({ type: \"toolcall_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\t\ttype: \"toolcall_delta\",\n\t\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\t\tdelta: JSON.stringify(toolCall.arguments),\n\t\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\t\ttype: \"toolcall_end\",\n\t\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\t\ttoolCall,\n\t\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\tif (candidate?.finishReason) {\n\t\t\t\t\t\t\t\toutput.stopReason = mapStopReasonString(candidate.finishReason);\n\t\t\t\t\t\t\t\tif (output.content.some((b) => b.type === \"toolCall\")) {\n\t\t\t\t\t\t\t\t\toutput.stopReason = \"toolUse\";\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\tif (responseData.usageMetadata) {\n\t\t\t\t\t\t\t\t// promptTokenCount includes cachedContentTokenCount, so subtract to get fresh input\n\t\t\t\t\t\t\t\tconst promptTokens = responseData.usageMetadata.promptTokenCount || 0;\n\t\t\t\t\t\t\t\tconst cacheReadTokens = responseData.usageMetadata.cachedContentTokenCount || 0;\n\t\t\t\t\t\t\t\toutput.usage = {\n\t\t\t\t\t\t\t\t\tinput: promptTokens - cacheReadTokens,\n\t\t\t\t\t\t\t\t\toutput:\n\t\t\t\t\t\t\t\t\t\t(responseData.usageMetadata.candidatesTokenCount || 0) +\n\t\t\t\t\t\t\t\t\t\t(responseData.usageMetadata.thoughtsTokenCount || 0),\n\t\t\t\t\t\t\t\t\tcacheRead: cacheReadTokens,\n\t\t\t\t\t\t\t\t\tcacheWrite: 0,\n\t\t\t\t\t\t\t\t\ttotalTokens: responseData.usageMetadata.totalTokenCount || 0,\n\t\t\t\t\t\t\t\t\tcost: {\n\t\t\t\t\t\t\t\t\t\tinput: 0,\n\t\t\t\t\t\t\t\t\t\toutput: 0,\n\t\t\t\t\t\t\t\t\t\tcacheRead: 0,\n\t\t\t\t\t\t\t\t\t\tcacheWrite: 0,\n\t\t\t\t\t\t\t\t\t\ttotal: 0,\n\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t};\n\t\t\t\t\t\t\t\tcalculateCost(model, output.usage);\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t} finally {\n\t\t\t\t\toptions?.signal?.removeEventListener(\"abort\", abortHandler);\n\t\t\t\t}\n\n\t\t\t\tif (currentBlock) {\n\t\t\t\t\tif (currentBlock.type === \"text\") {\n\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\tcontent: currentBlock.text,\n\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t});\n\t\t\t\t\t} else {\n\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\tcontent: currentBlock.thinking,\n\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t});\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\treturn hasContent;\n\t\t\t};\n\n\t\t\tlet receivedContent = false;\n\t\t\tlet currentResponse = response;\n\n\t\t\tfor (let emptyAttempt = 0; emptyAttempt <= MAX_EMPTY_STREAM_RETRIES; emptyAttempt++) {\n\t\t\t\tif (options?.signal?.aborted) {\n\t\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t\t}\n\n\t\t\t\tif (emptyAttempt > 0) {\n\t\t\t\t\tconst backoffMs = EMPTY_STREAM_BASE_DELAY_MS * 2 ** (emptyAttempt - 1);\n\t\t\t\t\tawait sleep(backoffMs, options?.signal);\n\n\t\t\t\t\tif (!requestUrl) {\n\t\t\t\t\t\tthrow new Error(\"Missing request URL\");\n\t\t\t\t\t}\n\n\t\t\t\t\tcurrentResponse = await fetch(requestUrl, {\n\t\t\t\t\t\tmethod: \"POST\",\n\t\t\t\t\t\theaders: requestHeaders,\n\t\t\t\t\t\tbody: requestBodyJson,\n\t\t\t\t\t\tsignal: options?.signal,\n\t\t\t\t\t});\n\t\t\t\t\tawait options?.onResponse?.(\n\t\t\t\t\t\t{ status: currentResponse.status, headers: headersToRecord(currentResponse.headers) },\n\t\t\t\t\t\tmodel,\n\t\t\t\t\t);\n\n\t\t\t\t\tif (!currentResponse.ok) {\n\t\t\t\t\t\tconst retryErrorText = await currentResponse.text();\n\t\t\t\t\t\tthrow new Error(`Cloud Code Assist API error (${currentResponse.status}): ${retryErrorText}`);\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\tconst streamed = await streamResponse(currentResponse);\n\t\t\t\tif (streamed) {\n\t\t\t\t\treceivedContent = true;\n\t\t\t\t\tbreak;\n\t\t\t\t}\n\n\t\t\t\tif (emptyAttempt < MAX_EMPTY_STREAM_RETRIES) {\n\t\t\t\t\tresetOutput();\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (!receivedContent) {\n\t\t\t\tthrow new Error(\"Cloud Code Assist API returned an empty response\");\n\t\t\t}\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(\"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) {\n\t\t\t\tif (\"index\" in block) {\n\t\t\t\t\tdelete (block as { index?: number }).index;\n\t\t\t\t}\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimpleGoogleGeminiCli: StreamFunction<\"google-gemini-cli\", SimpleStreamOptions> = (\n\tmodel: Model<\"google-gemini-cli\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey;\n\tif (!apiKey) {\n\t\tthrow new Error(\"Google Cloud Code Assist requires OAuth authentication. Use /login to authenticate.\");\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tif (!options?.reasoning) {\n\t\treturn streamGoogleGeminiCli(model, context, {\n\t\t\t...base,\n\t\t\tthinking: { enabled: false },\n\t\t} satisfies GoogleGeminiCliOptions);\n\t}\n\n\tconst effort = clampReasoning(options.reasoning)!;\n\tif (isGemini3Model(model.id)) {\n\t\treturn streamGoogleGeminiCli(model, context, {\n\t\t\t...base,\n\t\t\tthinking: {\n\t\t\t\tenabled: true,\n\t\t\t\tlevel: getGeminiCliThinkingLevel(effort, model.id),\n\t\t\t},\n\t\t} satisfies GoogleGeminiCliOptions);\n\t}\n\n\tconst defaultBudgets: ThinkingBudgets = {\n\t\tminimal: 1024,\n\t\tlow: 2048,\n\t\tmedium: 8192,\n\t\thigh: 16384,\n\t};\n\tconst budgets = { ...defaultBudgets, ...options.thinkingBudgets };\n\n\tconst minOutputTokens = 1024;\n\tlet thinkingBudget = budgets[effort]!;\n\tconst maxTokens = Math.min((base.maxTokens || 0) + thinkingBudget, model.maxTokens);\n\n\tif (maxTokens <= thinkingBudget) {\n\t\tthinkingBudget = Math.max(0, maxTokens - minOutputTokens);\n\t}\n\n\treturn streamGoogleGeminiCli(model, context, {\n\t\t...base,\n\t\tmaxTokens,\n\t\tthinking: {\n\t\t\tenabled: true,\n\t\t\tbudgetTokens: thinkingBudget,\n\t\t},\n\t} satisfies GoogleGeminiCliOptions);\n};\n\nexport function buildRequest(\n\tmodel: Model<\"google-gemini-cli\">,\n\tcontext: Context,\n\tprojectId: string,\n\toptions: GoogleGeminiCliOptions = {},\n\tisAntigravity = false,\n): CloudCodeAssistRequest {\n\tconst contents = convertMessages(model, context);\n\n\tconst generationConfig: CloudCodeAssistRequest[\"request\"][\"generationConfig\"] = {};\n\tif (options.temperature !== undefined) {\n\t\tgenerationConfig.temperature = options.temperature;\n\t}\n\tif (options.maxTokens !== undefined) {\n\t\tgenerationConfig.maxOutputTokens = options.maxTokens;\n\t}\n\n\t// Thinking config\n\tif (options.thinking?.enabled && model.reasoning) {\n\t\tgenerationConfig.thinkingConfig = {\n\t\t\tincludeThoughts: true,\n\t\t};\n\t\t// Gemini 3 models use thinkingLevel, older models use thinkingBudget\n\t\tif (options.thinking.level !== undefined) {\n\t\t\t// Cast to any since our GoogleThinkingLevel mirrors Google's ThinkingLevel enum values\n\t\t\tgenerationConfig.thinkingConfig.thinkingLevel = options.thinking.level as any;\n\t\t} else if (options.thinking.budgetTokens !== undefined) {\n\t\t\tgenerationConfig.thinkingConfig.thinkingBudget = options.thinking.budgetTokens;\n\t\t}\n\t} else if (model.reasoning && options.thinking && !options.thinking.enabled) {\n\t\tgenerationConfig.thinkingConfig = getDisabledThinkingConfig(model.id);\n\t}\n\n\tconst request: CloudCodeAssistRequest[\"request\"] = {\n\t\tcontents,\n\t};\n\n\trequest.sessionId = options.sessionId;\n\n\t// System instruction must be object with parts, not plain string\n\tif (context.systemPrompt) {\n\t\trequest.systemInstruction = {\n\t\t\tparts: [{ text: sanitizeSurrogates(context.systemPrompt) }],\n\t\t};\n\t}\n\n\tif (Object.keys(generationConfig).length > 0) {\n\t\trequest.generationConfig = generationConfig;\n\t}\n\n\tif (context.tools && context.tools.length > 0) {\n\t\t// Claude models on Cloud Code Assist need the legacy `parameters` field;\n\t\t// the API translates it into Anthropic's `input_schema`.\n\t\tconst useParameters = model.id.startsWith(\"claude-\");\n\t\trequest.tools = convertTools(context.tools, useParameters);\n\t\tif (options.toolChoice) {\n\t\t\trequest.toolConfig = {\n\t\t\t\tfunctionCallingConfig: {\n\t\t\t\t\tmode: mapToolChoice(options.toolChoice),\n\t\t\t\t},\n\t\t\t};\n\t\t}\n\t}\n\n\tif (isAntigravity) {\n\t\tconst existingParts = request.systemInstruction?.parts ?? [];\n\t\trequest.systemInstruction = {\n\t\t\trole: \"user\",\n\t\t\tparts: [\n\t\t\t\t{ text: ANTIGRAVITY_SYSTEM_INSTRUCTION },\n\t\t\t\t{ text: `Please ignore following [ignore]${ANTIGRAVITY_SYSTEM_INSTRUCTION}[/ignore]` },\n\t\t\t\t...existingParts,\n\t\t\t],\n\t\t};\n\t}\n\n\treturn {\n\t\tproject: projectId,\n\t\tmodel: model.id,\n\t\trequest,\n\t\t...(isAntigravity ? { requestType: \"agent\" } : {}),\n\t\tuserAgent: isAntigravity ? \"antigravity\" : \"pi-coding-agent\",\n\t\trequestId: `${isAntigravity ? \"agent\" : \"pi\"}-${Date.now()}-${Math.random().toString(36).slice(2, 11)}`,\n\t};\n}\n\ntype ClampedThinkingLevel = Exclude<ThinkingLevel, \"xhigh\">;\n\nfunction getDisabledThinkingConfig(modelId: string): ThinkingConfig {\n\t// Google docs: Gemini 3.1 Pro cannot disable thinking, and Gemini 3 Flash / Flash-Lite\n\t// do not support full thinking-off either. For Gemini 3 models, use the lowest supported\n\t// thinkingLevel without includeThoughts so hidden thinking remains invisible to pi.\n\tif (isGemini3ProModel(modelId)) {\n\t\treturn { thinkingLevel: \"LOW\" as any };\n\t}\n\tif (isGemini3FlashModel(modelId)) {\n\t\treturn { thinkingLevel: \"MINIMAL\" as any };\n\t}\n\n\t// Gemini 2.x supports disabling via thinkingBudget = 0.\n\treturn { thinkingBudget: 0 };\n}\n\nfunction getGeminiCliThinkingLevel(effort: ClampedThinkingLevel, modelId: string): GoogleThinkingLevel {\n\tif (isGemini3ProModel(modelId)) {\n\t\tswitch (effort) {\n\t\t\tcase \"minimal\":\n\t\t\tcase \"low\":\n\t\t\t\treturn \"LOW\";\n\t\t\tcase \"medium\":\n\t\t\tcase \"high\":\n\t\t\t\treturn \"HIGH\";\n\t\t}\n\t}\n\tswitch (effort) {\n\t\tcase \"minimal\":\n\t\t\treturn \"MINIMAL\";\n\t\tcase \"low\":\n\t\t\treturn \"LOW\";\n\t\tcase \"medium\":\n\t\t\treturn \"MEDIUM\";\n\t\tcase \"high\":\n\t\t\treturn \"HIGH\";\n\t}\n}\n"]}