@standardagents/deepinfra 0.0.1-dev.fffff

package/README.md

@@ -0,0 +1,26 @@
+# @standardagents/deepinfra
+
+First-party [DeepInfra](https://deepinfra.com/) provider for Standard Agents.
+
+```bash
+pnpm add @standardagents/deepinfra
+```
+
+```ts
+import { defineModel } from '@standardagents/builder';
+import { deepinfra } from '@standardagents/deepinfra';
+
+export default defineModel({
+  name: 'deepinfra-chat',
+  provider: deepinfra,
+  model: 'deepseek-ai/DeepSeek-V3',
+});
+```
+
+Set `DEEPINFRA_API_KEY` for direct BYOK requests. The provider uses DeepInfra's OpenAI-compatible Chat Completions endpoint, discovers chat models and per-million-token prices from the live public catalog, supports streaming, images, tools, strict structured output, reasoning controls, prompt caching options, and request inspection, and preserves DeepInfra's response-level `usage.estimated_cost` as `usage.cost`.
+
+Model discovery also resolves the model namespace (for example `deepseek-ai`, `meta-llama`, or `Qwen`) to bundled local lab artwork, so Builder model lists show the model creator rather than repeating the DeepInfra provider mark.
+
+Connection slots are `apiKey`, `baseUrl` (default `https://api.deepinfra.com/v1/openai`), `modelsUrl` (default `https://api.deepinfra.com/v1/models`), and `timeout`.
+
+See the [DeepInfra provider guide](https://docs.standardagentbuilder.com/providers/deepinfra) and [DeepInfra API docs](https://docs.deepinfra.com/chat/overview).

package/dist/index.d.ts

@@ -0,0 +1,93 @@
+import { LLMProviderInterface, ProviderFactoryConfig, ModelCapabilities, ProviderModelInfo, ProviderRequest, ProviderResponse, ProviderStreamChunk, InspectedRequest, ProviderUsagePricing, ProviderUsage, ProviderFactoryWithOptions } from '@standardagents/spec';
+import { z } from 'zod';
+
+/** DeepInfra-specific and OpenAI-compatible Chat Completions options. */
+declare const deepinfraProviderOptions: z.ZodObject<{
+    n: z.ZodOptional<z.ZodNumber>;
+    seed: z.ZodOptional<z.ZodNumber>;
+    frequency_penalty: z.ZodOptional<z.ZodNumber>;
+    presence_penalty: z.ZodOptional<z.ZodNumber>;
+    repetition_penalty: z.ZodOptional<z.ZodNumber>;
+    top_k: z.ZodOptional<z.ZodNumber>;
+    min_p: z.ZodOptional<z.ZodNumber>;
+    logit_bias: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodNumber>>;
+    logprobs: z.ZodOptional<z.ZodBoolean>;
+    top_logprobs: z.ZodOptional<z.ZodNumber>;
+    stop_token_ids: z.ZodOptional<z.ZodArray<z.ZodNumber>>;
+    reasoning_effort: z.ZodOptional<z.ZodEnum<{
+        none: "none";
+        low: "low";
+        medium: "medium";
+        high: "high";
+    }>>;
+    reasoning: z.ZodOptional<z.ZodObject<{
+        enabled: z.ZodOptional<z.ZodBoolean>;
+        effort: z.ZodOptional<z.ZodEnum<{
+            none: "none";
+            low: "low";
+            medium: "medium";
+            high: "high";
+        }>>;
+    }, z.core.$loose>>;
+    service_tier: z.ZodOptional<z.ZodEnum<{
+        default: "default";
+        priority: "priority";
+        flex: "flex";
+    }>>;
+    prompt_cache_key: z.ZodOptional<z.ZodString>;
+}, z.core.$loose>;
+type DeepInfraProviderOptions = z.infer<typeof deepinfraProviderOptions>;
+
+interface DeepInfraModelInfo {
+    id: string;
+    object?: string;
+    created?: number;
+    owned_by?: string;
+    metadata?: {
+        description?: string;
+        context_length?: number;
+        max_tokens?: number;
+        pricing?: {
+            input_tokens?: number;
+            output_tokens?: number;
+            cache_read_tokens?: number;
+        };
+        tags?: string[];
+    };
+}
+declare function normalizeDeepInfraPricePerMillion(value: unknown): number | undefined;
+declare function deepinfraUsagePricingFromModel(model: DeepInfraModelInfo | undefined): ProviderUsagePricing | undefined;
+declare function calculateDeepInfraUsageCost(usage: ProviderUsage, pricing?: ProviderUsagePricing): number | undefined;
+declare class DeepInfraProvider implements LLMProviderInterface {
+    readonly name = "deepinfra";
+    readonly specificationVersion: "1";
+    private config;
+    private modelsCache;
+    private modelsCacheTime;
+    private static readonly CACHE_TTL;
+    constructor(config: ProviderFactoryConfig);
+    supportsModel(_modelId: string): boolean;
+    getIcon(modelId?: string): string;
+    private baseUrl;
+    private modelsUrl;
+    private fetch;
+    private headers;
+    private requestGuard;
+    private fetchModelsWithCache;
+    private mapModelCapabilities;
+    getModelCapabilities(modelId: string): Promise<ModelCapabilities | null>;
+    private mapToProviderModelInfo;
+    getModels(filter?: string): Promise<ProviderModelInfo[]>;
+    private getUsagePricing;
+    generate(request: ProviderRequest): Promise<ProviderResponse>;
+    stream(request: ProviderRequest): Promise<AsyncIterable<ProviderStreamChunk>>;
+    private toHttpProviderErrorFromResponse;
+    private toProviderError;
+    private toHttpProviderError;
+    inspectRequest(request: ProviderRequest): Promise<InspectedRequest>;
+}
+
+/** DeepInfra provider factory backed by its OpenAI-compatible chat API and live model catalog. */
+declare const deepinfra: ProviderFactoryWithOptions<typeof deepinfraProviderOptions>;
+
+export { type DeepInfraModelInfo, DeepInfraProvider, type DeepInfraProviderOptions, calculateDeepInfraUsageCost, deepinfra, deepinfraProviderOptions, deepinfraUsagePricingFromModel, normalizeDeepInfraPricePerMillion };