@truefoundry/ai-gateway-sdk 0.0.0-stage → 0.0.2

package/dist/entities/Model.d.ts

@@ -0,0 +1,105 @@
+import { GuardrailConfigIntegrations, ModelProviderAccount, OCRCostMetric, PerMillionCharactersCostMetric, PerMinuteOfAudioCostMetric, PerSecondOfVideoCostMetric, PerThousandEmbeddingTokensCostMetric, PerThousandTokensCostMetric, PublicCostMetric, VirtualModel, VirtualModelProviderAccount } from '../autogen/models.js';
+import { Provider } from './Provider.js';
+export declare enum ModelType {
+    chat = "chat",
+    responses = "responses",
+    completion = "completion",
+    embedding = "embedding",
+    rerank = "rerank",
+    realtime = "realtime",
+    audio_transcription = "audio_transcription",
+    audio_translation = "audio_translation",
+    text_to_speech = "text_to_speech",
+    moderation = "moderation",
+    image = "image",
+    ocr = "ocr",
+    video = "video",
+    proxy = "proxy"
+}
+export declare enum ModelCostMetricType {
+    per_1000_tokens = "per_1000_tokens",
+    per_1000_embedding_tokens = "per_1000_embedding_tokens",
+    per_million_characters = "per_million_characters",
+    per_minute_of_audio = "per_minute_of_audio",
+    per_1000_pages = "per_1000_pages",
+    per_second_of_video = "per_second_of_video",
+    public_cost = "public_cost"
+}
+export declare enum ModelServerType {
+    openai_compatible = "openai-compatible",
+    vllm_openai = "vllm-openai",
+    tei = "tei",
+    infinity = "infinity",
+    nemo_retriever = "nemo-retriever",
+    cohere_rerank_v2 = "cohere-rerank-v2",
+    comfyui = "comfyui"
+}
+export interface PerThousandEmbeddingTokensCostMetricInterface extends PerThousandEmbeddingTokensCostMetric {
+    metric: ModelCostMetricType.per_1000_embedding_tokens;
+}
+export interface PerThousandTokensCostMetricInterface extends PerThousandTokensCostMetric {
+    metric: ModelCostMetricType.per_1000_tokens;
+}
+export interface PublicCostMetricInterface extends PublicCostMetric {
+    metric: ModelCostMetricType.public_cost;
+}
+export interface PerMillionCharactersCostMetricInterface extends PerMillionCharactersCostMetric {
+    metric: ModelCostMetricType.per_million_characters;
+}
+export interface OCRCostMetricInterface extends OCRCostMetric {
+    metric: ModelCostMetricType.per_1000_pages;
+}
+export interface PerMinuteOfAudioCostMetricInterface extends PerMinuteOfAudioCostMetric {
+    metric: ModelCostMetricType.per_minute_of_audio;
+}
+export interface PerSecondOfVideoCostMetricInterface extends PerSecondOfVideoCostMetric {
+    metric: ModelCostMetricType.per_second_of_video;
+}
+type ExtractIntegrationsElement<T> = T extends {
+    integrations: readonly (infer U)[];
+} ? U : T extends {
+    integrations?: readonly (infer U)[];
+} ? U : T extends {
+    integrations: (infer U)[];
+} ? U : T extends {
+    integrations?: (infer U)[];
+} ? U : never;
+export type ProviderModelConfig = ExtractIntegrationsElement<Exclude<ModelProviderAccount, VirtualModelProviderAccount>>;
+export type ModelConfig = ProviderModelConfig | VirtualModel;
+type ProviderAccount = {
+    id?: string;
+    fqn: string;
+    name: string;
+    provider: Provider;
+    tenantName: string;
+};
+export declare class Model {
+    id?: string;
+    fqn: string;
+    name: string;
+    manifest: ModelConfig;
+    providerAccount: ProviderAccount;
+    createdBy: string;
+    constructor(input: {
+        id?: string;
+        fqn: string;
+        name: string;
+        manifest: ProviderModelConfig;
+        providerAccount: ProviderAccount;
+        createdBy: string;
+    });
+}
+export interface ProviderWiseModels {
+    [provider: string]: {
+        [providerAccountName: string]: Model[];
+    };
+}
+export interface GuardrailConfigIntegrationEntity {
+    id?: string;
+    fqn: string;
+    name: string;
+    manifest: GuardrailConfigIntegrations;
+    providerAccount: ProviderAccount;
+    createdBy: string;
+}
+export {};

package/dist/entities/Model.js

@@ -0,0 +1,55 @@
+// TODO: Add messages when we have a messages modelType
+export var ModelType;
+(function (ModelType) {
+    ModelType["chat"] = "chat";
+    ModelType["responses"] = "responses";
+    ModelType["completion"] = "completion";
+    ModelType["embedding"] = "embedding";
+    ModelType["rerank"] = "rerank";
+    ModelType["realtime"] = "realtime";
+    ModelType["audio_transcription"] = "audio_transcription";
+    ModelType["audio_translation"] = "audio_translation";
+    ModelType["text_to_speech"] = "text_to_speech";
+    ModelType["moderation"] = "moderation";
+    ModelType["image"] = "image";
+    ModelType["ocr"] = "ocr";
+    ModelType["video"] = "video";
+    ModelType["proxy"] = "proxy";
+})(ModelType || (ModelType = {}));
+export var ModelCostMetricType;
+(function (ModelCostMetricType) {
+    ModelCostMetricType["per_1000_tokens"] = "per_1000_tokens";
+    ModelCostMetricType["per_1000_embedding_tokens"] = "per_1000_embedding_tokens";
+    ModelCostMetricType["per_million_characters"] = "per_million_characters";
+    ModelCostMetricType["per_minute_of_audio"] = "per_minute_of_audio";
+    ModelCostMetricType["per_1000_pages"] = "per_1000_pages";
+    ModelCostMetricType["per_second_of_video"] = "per_second_of_video";
+    ModelCostMetricType["public_cost"] = "public_cost";
+})(ModelCostMetricType || (ModelCostMetricType = {}));
+export var ModelServerType;
+(function (ModelServerType) {
+    ModelServerType["openai_compatible"] = "openai-compatible";
+    ModelServerType["vllm_openai"] = "vllm-openai";
+    ModelServerType["tei"] = "tei";
+    ModelServerType["infinity"] = "infinity";
+    ModelServerType["nemo_retriever"] = "nemo-retriever";
+    ModelServerType["cohere_rerank_v2"] = "cohere-rerank-v2";
+    ModelServerType["comfyui"] = "comfyui";
+})(ModelServerType || (ModelServerType = {}));
+export class Model {
+    id;
+    fqn;
+    name;
+    manifest;
+    providerAccount;
+    createdBy;
+    constructor(input) {
+        this.id = input.id;
+        this.name = input.name;
+        this.providerAccount = input.providerAccount;
+        this.manifest = input.manifest;
+        this.fqn = input.fqn;
+        this.createdBy = input.createdBy;
+    }
+}
+//# sourceMappingURL=Model.js.map

package/dist/entities/Model.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"Model.js","sourceRoot":"","sources":["../../src/entities/Model.ts"],"names":[],"mappings":"AAeA,uDAAuD;AACvD,MAAM,CAAN,IAAY,SAeX;AAfD,WAAY,SAAS;IACnB,0BAAa,CAAA;IACb,oCAAuB,CAAA;IACvB,sCAAyB,CAAA;IACzB,oCAAuB,CAAA;IACvB,8BAAiB,CAAA;IACjB,kCAAqB,CAAA;IACrB,wDAA2C,CAAA;IAC3C,oDAAuC,CAAA;IACvC,8CAAiC,CAAA;IACjC,sCAAyB,CAAA;IACzB,4BAAe,CAAA;IACf,wBAAW,CAAA;IACX,4BAAe,CAAA;IACf,4BAAe,CAAA;AACjB,CAAC,EAfW,SAAS,KAAT,SAAS,QAepB;AAED,MAAM,CAAN,IAAY,mBAQX;AARD,WAAY,mBAAmB;IAC7B,0DAAmC,CAAA;IACnC,8EAAuD,CAAA;IACvD,wEAAiD,CAAA;IACjD,kEAA2C,CAAA;IAC3C,wDAAiC,CAAA;IACjC,kEAA2C,CAAA;IAC3C,kDAA2B,CAAA;AAC7B,CAAC,EARW,mBAAmB,KAAnB,mBAAmB,QAQ9B;AAED,MAAM,CAAN,IAAY,eAQX;AARD,WAAY,eAAe;IACzB,0DAAuC,CAAA;IACvC,8CAA2B,CAAA;IAC3B,8BAAW,CAAA;IACX,wCAAqB,CAAA;IACrB,oDAAiC,CAAA;IACjC,wDAAqC,CAAA;IACrC,sCAAmB,CAAA;AACrB,CAAC,EARW,eAAe,KAAf,eAAe,QAQ1B;AAuDD,MAAM,OAAO,KAAK;IAChB,EAAE,CAAU;IACZ,GAAG,CAAS;IACZ,IAAI,CAAS;IACb,QAAQ,CAAc;IACtB,eAAe,CAAkB;IACjC,SAAS,CAAS;IAElB,YAAY,KAOX;QACC,IAAI,CAAC,EAAE,GAAG,KAAK,CAAC,EAAE,CAAC;QACnB,IAAI,CAAC,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC;QACvB,IAAI,CAAC,eAAe,GAAG,KAAK,CAAC,eAAe,CAAC;QAC7C,IAAI,CAAC,QAAQ,GAAG,KAAK,CAAC,QAAQ,CAAC;QAC/B,IAAI,CAAC,GAAG,GAAG,KAAK,CAAC,GAAG,CAAC;QACrB,IAAI,CAAC,SAAS,GAAG,KAAK,CAAC,SAAS,CAAC;IACnC,CAAC;CACF"}

package/dist/entities/Provider.d.ts

@@ -0,0 +1,46 @@
+export declare enum Provider {
+    openai = "openai",
+    azure_openai = "azure",
+    cohere = "cohere",
+    truefoundry_public = "truefoundry",
+    custom = "custom",
+    aws = "aws",
+    ai21 = "ai21",
+    anthropic = "anthropic",
+    aws_claude_platform = "aws-claude-platform",
+    anyscale = "anyscale",
+    deepinfra = "deepinfra",
+    google = "gcp",
+    groq = "groq",
+    mistral_ai = "mistral-ai",
+    nomic = "nomic",
+    ollama = "ollama",
+    palm = "palm",
+    perplexity_ai = "perplexity-ai",
+    together_ai = "together-ai",
+    aws_bedrock = "aws-bedrock",
+    aws_bedrock_mantle = "aws-bedrock-mantle",
+    azure_foundry = "azure-foundry",
+    microsoft_foundry = "microsoft-foundry",
+    vertex = "google-vertex",
+    gemini = "google-gemini",
+    azure_openai_v2 = "azure-openai",
+    databricks = "databricks",
+    aws_sagemaker = "aws-sagemaker",
+    sambanova = "sambanova",
+    cerebras = "cerebras",
+    wafer = "wafer",
+    typesafe = "typesafe",
+    openrouter = "openrouter",
+    snowflake_cortex = "snowflake-cortex",
+    cloudera = "cloudera",
+    xai = "xai",
+    baseten = "baseten",
+    cartesia = "cartesia",
+    deepgram = "deepgram",
+    virtual_model = "virtual-model",
+    elevenlabs = "elevenlabs",
+    custom_endpoint = "custom-endpoint",
+    smallest_ai = "smallest-ai",
+    truefoundry_models = "truefoundry-models"
+}

package/dist/entities/Provider.js

@@ -0,0 +1,48 @@
+export var Provider;
+(function (Provider) {
+    Provider["openai"] = "openai";
+    Provider["azure_openai"] = "azure";
+    Provider["cohere"] = "cohere";
+    Provider["truefoundry_public"] = "truefoundry";
+    Provider["custom"] = "custom";
+    Provider["aws"] = "aws";
+    Provider["ai21"] = "ai21";
+    Provider["anthropic"] = "anthropic";
+    Provider["aws_claude_platform"] = "aws-claude-platform";
+    Provider["anyscale"] = "anyscale";
+    Provider["deepinfra"] = "deepinfra";
+    Provider["google"] = "gcp";
+    Provider["groq"] = "groq";
+    Provider["mistral_ai"] = "mistral-ai";
+    Provider["nomic"] = "nomic";
+    Provider["ollama"] = "ollama";
+    Provider["palm"] = "palm";
+    Provider["perplexity_ai"] = "perplexity-ai";
+    Provider["together_ai"] = "together-ai";
+    Provider["aws_bedrock"] = "aws-bedrock";
+    Provider["aws_bedrock_mantle"] = "aws-bedrock-mantle";
+    Provider["azure_foundry"] = "azure-foundry";
+    Provider["microsoft_foundry"] = "microsoft-foundry";
+    Provider["vertex"] = "google-vertex";
+    Provider["gemini"] = "google-gemini";
+    Provider["azure_openai_v2"] = "azure-openai";
+    Provider["databricks"] = "databricks";
+    Provider["aws_sagemaker"] = "aws-sagemaker";
+    Provider["sambanova"] = "sambanova";
+    Provider["cerebras"] = "cerebras";
+    Provider["wafer"] = "wafer";
+    Provider["typesafe"] = "typesafe";
+    Provider["openrouter"] = "openrouter";
+    Provider["snowflake_cortex"] = "snowflake-cortex";
+    Provider["cloudera"] = "cloudera";
+    Provider["xai"] = "xai";
+    Provider["baseten"] = "baseten";
+    Provider["cartesia"] = "cartesia";
+    Provider["deepgram"] = "deepgram";
+    Provider["virtual_model"] = "virtual-model";
+    Provider["elevenlabs"] = "elevenlabs";
+    Provider["custom_endpoint"] = "custom-endpoint";
+    Provider["smallest_ai"] = "smallest-ai";
+    Provider["truefoundry_models"] = "truefoundry-models";
+})(Provider || (Provider = {}));
+//# sourceMappingURL=Provider.js.map

package/dist/entities/Provider.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"Provider.js","sourceRoot":"","sources":["../../src/entities/Provider.ts"],"names":[],"mappings":"AAAA,MAAM,CAAN,IAAY,QA6CX;AA7CD,WAAY,QAAQ;IAClB,6BAAiB,CAAA;IACjB,kCAAsB,CAAA;IACtB,6BAAiB,CAAA;IACjB,8CAAkC,CAAA;IAClC,6BAAiB,CAAA;IACjB,uBAAW,CAAA;IACX,yBAAa,CAAA;IACb,mCAAuB,CAAA;IACvB,uDAA2C,CAAA;IAC3C,iCAAqB,CAAA;IACrB,mCAAuB,CAAA;IACvB,0BAAc,CAAA;IACd,yBAAa,CAAA;IACb,qCAAyB,CAAA;IACzB,2BAAe,CAAA;IACf,6BAAiB,CAAA;IACjB,yBAAa,CAAA;IACb,2CAA+B,CAAA;IAC/B,uCAA2B,CAAA;IAC3B,uCAA2B,CAAA;IAC3B,qDAAyC,CAAA;IACzC,2CAA+B,CAAA;IAC/B,mDAAuC,CAAA;IACvC,oCAAwB,CAAA;IACxB,oCAAwB,CAAA;IACxB,4CAAgC,CAAA;IAChC,qCAAyB,CAAA;IACzB,2CAA+B,CAAA;IAC/B,mCAAuB,CAAA;IACvB,iCAAqB,CAAA;IACrB,2BAAe,CAAA;IACf,iCAAqB,CAAA;IACrB,qCAAyB,CAAA;IACzB,iDAAqC,CAAA;IACrC,iCAAqB,CAAA;IACrB,uBAAW,CAAA;IACX,+BAAmB,CAAA;IACnB,iCAAqB,CAAA;IACrB,iCAAqB,CAAA;IACrB,2CAA+B,CAAA;IAC/B,qCAAyB,CAAA;IACzB,+CAAmC,CAAA;IACnC,uCAA2B,CAAA;IAC3B,qDAAyC,CAAA;AAC3C,CAAC,EA7CW,QAAQ,KAAR,QAAQ,QA6CnB"}

package/dist/entities/ProviderAccount.d.ts

@@ -0,0 +1,23 @@
+import type { InfraProviderAccount, ProviderAccounts, ProviderIntegrations } from '../autogen/models.js';
+import { Model } from './Model.js';
+import { Provider } from './Provider.js';
+export type ProviderAccountConfig = Exclude<ProviderAccounts, InfraProviderAccount>;
+export interface ProviderAccountEntity {
+    id?: string;
+    fqn: string;
+    name: string;
+    provider: Provider;
+    tenantName: string;
+    manifest: ProviderAccountConfig;
+    integrations?: ProviderIntegrationsEntity[] | Model[];
+    /** Absent on the provider accounts a V2 snapshot publishes, which are entity rows without it. */
+    createdBy?: string;
+}
+export interface ProviderIntegrationsEntity {
+    id?: string;
+    fqn: string;
+    name: string;
+    manifest: ProviderIntegrations;
+    providerAccount: ProviderAccountEntity;
+    createdBy: string;
+}

package/dist/entities/ProviderAccount.js

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=ProviderAccount.js.map

package/dist/entities/ProviderAccount.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"ProviderAccount.js","sourceRoot":"","sources":["../../src/entities/ProviderAccount.ts"],"names":[],"mappings":""}

package/dist/errors.d.ts

@@ -0,0 +1,190 @@
+import { HTTPException } from 'hono/http-exception';
+import { TFY_ERROR_TYPES } from './constants.js';
+export declare class UnrecoverableError extends Error {
+    constructor(message: string | undefined);
+}
+export declare class GatewayUnexpectedError extends Error {
+    cause?: Error | undefined;
+    constructor(message: string, cause?: Error | undefined);
+}
+export declare class BaseError extends HTTPException {
+    statusCode: number;
+    message: string;
+    options?: ErrorOptions;
+    constructor(statusCode: number, message: string, options?: ErrorOptions & {
+        cause?: any;
+    });
+    toJSON(): {
+        name: string;
+        message: string;
+        statusCode: number;
+    };
+    toString(): string;
+}
+export declare class RateLimitError extends BaseError {
+    statusCode: number;
+    message: string;
+    readonly errorType: typeof TFY_ERROR_TYPES.RATE_LIMIT;
+    constructor(message: string, options?: ErrorOptions & {
+        cause?: any;
+    });
+}
+export declare class BudgetLimitError extends BaseError {
+    statusCode: number;
+    message: string;
+    readonly errorType: typeof TFY_ERROR_TYPES.BUDGET_LIMIT;
+    constructor(message: string, options?: ErrorOptions & {
+        cause?: any;
+    });
+}
+export declare class AuthenticationError extends BaseError {
+    statusCode: number;
+    message: string;
+    constructor(message: string, options?: ErrorOptions & {
+        cause?: any;
+    });
+}
+export declare class ValidationError extends BaseError {
+    statusCode: number;
+    message: string;
+    cause?: any;
+    constructor(message: string, options?: ErrorOptions & {
+        cause?: any;
+    });
+}
+export declare class AuthorizationError extends BaseError {
+    statusCode: number;
+    message: string;
+    options?: ErrorOptions;
+    cause?: any;
+    errorType: typeof TFY_ERROR_TYPES.TFY_FORBIDDEN | typeof TFY_ERROR_TYPES.MODEL_NOT_FOUND;
+    constructor(message: string, options?: ErrorOptions & {
+        cause?: any;
+        statusCode?: number;
+        errorType?: typeof TFY_ERROR_TYPES.TFY_FORBIDDEN | typeof TFY_ERROR_TYPES.MODEL_NOT_FOUND;
+    });
+}
+export declare class CredentialResolutionError extends BaseError {
+    constructor(message: string, options?: ErrorOptions & {
+        cause?: any;
+    });
+}
+/**
+ * Raised for stateful Responses-API requests on a virtual model that cannot be served:
+ * the pinned origin backend is unavailable, no longer configured, or the origin cannot
+ * resume stored state (e.g. a translated backend). 409 Conflict — the request conflicts
+ * with the conversation's bound backend; the gateway must NOT reroute (that would break the
+ * conversation), so it surfaces a clear, actionable error instead.
+ */
+export declare class ResponsesPinningError extends BaseError {
+    statusCode: number;
+    constructor(message: string, options?: ErrorOptions & {
+        cause?: any;
+    });
+}
+/**
+ * Raised when a virtual-model Responses API request would be served by a target that does not
+ * support the Responses API natively (it would be translated to Chat Completions). Translation
+ * drops previous_response_id and stores no server-side state, so a translated backend cannot back
+ * a stateful, origin-pinned Responses conversation. We reject instead of silently translating.
+ * 400 Bad Request — the status is intentionally NOT in DEFAULT_FALLBACK_STATUS_CODES so the
+ * load-balancing loop treats it as terminal and never falls back to a sibling target.
+ */
+export declare class ResponsesTranslationUnsupportedError extends BaseError {
+    statusCode: number;
+    constructor(message: string, options?: ErrorOptions & {
+        cause?: any;
+    });
+}
+/**
+ * Raised when a target's provider config has no messagesCountTokens entry. 404 is intentional: it IS in
+ * DEFAULT_FALLBACK_STATUS_CODES, so the load-balancing loop falls back to the next target instead of
+ * failing a mixed virtual model. Not an APIError, so it contributes no error metric or cooldown.
+ */
+export declare class CountTokensUnsupportedError extends BaseError {
+    statusCode: number;
+    constructor(message: string, options?: ErrorOptions & {
+        cause?: any;
+    });
+}
+export declare class APIError extends BaseError {
+    statusCode: number;
+    message: string;
+    promptTokenCount?: number;
+    requestBody?: string;
+    constructor(message: string, options: ErrorOptions & {
+        promptTokenCount?: number;
+        requestBody?: string;
+        statusCode: number;
+        cause?: any;
+    });
+}
+export declare class ProviderStreamError extends APIError {
+    providerErrorType: string | null;
+    providerErrorCode: string | null;
+    /** The provider's message alone. `message` folds in the type/code, which a span has nowhere else to put. */
+    providerMessage: string;
+    constructor(message: string, options: ErrorOptions & {
+        statusCode: number;
+        providerErrorType?: string | null;
+        providerErrorCode?: string | null;
+        providerMessage?: string;
+        cause?: any;
+    });
+}
+export declare class GuardrailsError extends BaseError {
+    readonly isGuardrailsError = true;
+    readonly guardrailChecks: any;
+    constructor(message: string, status?: number, guardrailChecks?: any);
+    static isGuardrailsError(error: any): error is GuardrailsError;
+}
+export declare class PromptHydrationError extends BaseError {
+    constructor(message: string);
+}
+export declare class TimeoutError extends BaseError {
+    statusCode: number;
+    message: string;
+    timeout: number;
+    constructor(message: string, timeout: number, options?: ErrorOptions & {
+        cause?: any;
+    });
+}
+export declare class RedisUnavailableError extends BaseError {
+    constructor();
+}
+export declare class TTFTTimeoutError extends BaseError {
+    statusCode: number;
+    message: string;
+    timeout: number;
+    constructor(message: string, timeout: number);
+}
+export declare class CacheError extends BaseError {
+    constructor(message: string, options?: ErrorOptions & {
+        cause?: any;
+    });
+}
+export declare class ResTranslationError extends BaseError {
+    cause?: Error;
+    constructor(cause?: Error);
+}
+export declare class ReqTranslationError extends BaseError {
+    cause?: Error;
+    constructor(cause?: Error);
+}
+export declare class RoutingConfigError extends BaseError {
+    constructor(message: string, statusCode?: number);
+}
+export declare class ReqBodyMalformedError extends BaseError {
+    constructor(cause?: unknown);
+}
+export declare class ClientDisconnectionError extends BaseError {
+    constructor(message?: string);
+}
+/** Maps an HTTP status code to the closest JSON-RPC error code. */
+export declare function jsonRpcCodeForStatus(status: number): number;
+/** One connect attempt and the upstream error it produced. */
+export interface ConnectStrategyFailure {
+    /** User-facing name of the attempt (strategy and transport), e.g. "Streamable HTTP auto negotiate". */
+    label: string;
+    error: string;
+}