@samesake/providers

Ready-made model-provider adapters for samesake's BYO seams. Generation, parsing, and reranking stay in this convenience package; Gemini and Voyage embedding factories are direct re-exports of the dual-form implementations in @samesake/embed.

import { createMatcher } from "@samesake/server";
import { geminiEmbedder, geminiGenerator, cohereReranker } from "@samesake/providers";

const matcher = createMatcher({
  databaseUrl: process.env.SAMESAKE_DATABASE_URL!,
  apiKey: process.env.SAMESAKE_API_KEY!,
  embed: geminiEmbedder(),        // EmbedFn — multimodal (text + image)
  generate: geminiGenerator(),    // GenerateFn — NLQ + enrichment
  rerank: cohereReranker(),       // RerankFn — optional second stage
});

Built-in adapters (zero dependencies, plain fetch)

| Provider | embed | generate | parse | rerank | key env var | |---|---|---|---|---|---| | Gemini | geminiEmbedder (multimodal, @samesake/embed) | geminiGenerator | geminiParser | — | GEMINI_API_KEY | | OpenAI | openaiEmbedder | openaiGenerator | openaiParser | — | OPENAI_API_KEY | | Voyage | voyageEmbedder (@samesake/embed) | — | — | voyageReranker | VOYAGE_API_KEY | | Cohere | cohereEmbedder | — | — | cohereReranker | COHERE_API_KEY |

The embedding aliases use the @samesake/embed factories and resolve API keys from the provider environment variables. Collection requests supply their model and dimension; the factories also accept their provider-specific defaults (dim and Gemini's taskType). The generation and reranking adapters accept the shared provider options:

geminiGenerator({
  apiKey: "...",        // default: the provider's env var
  model: "...",         // default model for generation
  baseUrl: "...",       // proxies / regional endpoints
  minIntervalMs: 120,   // space out calls during bulk indexing
  retries: 8,           // 429/5xx retries with exponential backoff (default 5)
});

Vercel AI SDK bridge (@samesake/providers/ai-sdk)

Already on the AI SDK? Wrap any of its model objects — the entire AI SDK provider ecosystem plugs into samesake without glue. ai is an optional peer dependency, only needed for this subpath.

import { google } from "@ai-sdk/google";
import { aiSdkEmbedder, aiSdkGenerator } from "@samesake/providers/ai-sdk";

createMatcher({
  embed: aiSdkEmbedder(google.textEmbedding("gemini-embedding-2"), {
    providerOptions: ({ dim, taskType }) => ({
      google: { outputDimensionality: dim, ...(taskType ? { taskType } : {}) },
    }),
  }),
  generate: aiSdkGenerator(google("gemini-2.5-flash-lite")),
});

Also available: aiSdkParser(model) and aiSdkReranker(model) (AI SDK v6 reranking models). Note the AI SDK's embed() is text-only — for image spaces (visual similarity) use the native multimodal geminiEmbedder or bring your own.