@llm-ports/adapter-vercel

Adapter for the Vercel AI SDK. Migration helper for users already using @ai-sdk/*. Implements LLMPort and EmbeddingsPort.

When to use this adapter

• You already have @ai-sdk/anthropic, @ai-sdk/openai, etc. wired into your project
• You want to add llm-ports (cost gating, fallback chains, capability factories) without rewriting the integration

For new projects, prefer the direct adapters (@llm-ports/adapter-anthropic, @llm-ports/adapter-openai). Fewer layers, full multimodal, full agent features.

Install

pnpm add @llm-ports/core @llm-ports/adapter-vercel ai @ai-sdk/anthropic

Configure

import { anthropic } from "@ai-sdk/anthropic";
import { openai } from "@ai-sdk/openai";
import { createRegistryFromEnv } from "@llm-ports/core";
import { createVercelAdapter } from "@llm-ports/adapter-vercel";

const registry = createRegistryFromEnv({
  adapters: {
    vercel: createVercelAdapter({
      models: {
        "claude-sonnet-4-6": anthropic("claude-sonnet-4-6"),
        "gpt-5": openai("gpt-5"),
      },
      embeddingModels: {
        "text-embedding-3-small": openai.textEmbeddingModel("text-embedding-3-small"),
      },
      pricing: {
        "claude-sonnet-4-6": { inputPer1M: 3, outputPer1M: 15 },
        "gpt-5": { inputPer1M: 2.5, outputPer1M: 10 },
        "text-embedding-3-small": { inputPer1M: 0, outputPer1M: 0, embeddingPer1M: 0.02 },
      },
    }),
  },
});

export const llm = registry.getPort();

You bring your own LanguageModel instances. The adapter routes LLMPort calls to Vercel's helpers (generateText, streamText, embed, embedMany).

Adapter options

interface VercelAdapterOptions {
  models?: Record<string, LanguageModel>;
  embeddingModels?: Record<string, EmbeddingModel<string>>;
  pricing: Record<string, ModelPricing>;     // REQUIRED
  validationStrategy?: ValidationStrategy;
}

pricing is required because the adapter has no built-in pricing table — models can come from any of @ai-sdk/anthropic, @ai-sdk/openai, @ai-sdk/google, etc. You supply the pricing for whatever you wire up.

Supported features

Feature	Status
generateText	✓
generateStructured (Zod schemas)	✓ (prompted JSON + retry-with-feedback)
streamText	✓
streamStructured	✓ (best-effort partial parse)
runAgent	Limited in v0.1: single-turn only; multi-turn tool use through Vercel's own agent loop comes in v0.2
generateEmbedding / generateEmbeddings	✓
Multimodal content blocks	partial (string conversion in v0.1; full multimodal in v0.2)

Limitations to know

These are also surfaced in the cross-package v0.1 status page along with their resolution paths.

• runAgent is single-turn in v0.1. Multi-step tool use through Vercel's own agent loop will land in v0.2 once the API surface is locked down. For multi-turn agents today, prefer the direct adapters.
• Multimodal is text-only in v0.1. Image and audio content blocks pass through as a stringified [image content] placeholder. Direct adapters support full multimodal.
• You bring your own pricing. No bundled table. Look up the rates for your chosen models from the provider's pricing page.
• No reasoning-model handling. The Vercel adapter does NOT apply the headroom multiplier the OpenAI adapter does. Calling against gpt-5-nano, o3, o3-mini, Cerebras gpt-oss-120b, or other reasoning models with a small maxOutputTokens (e.g. 20) reliably starves the model and returns empty text. Workaround: set maxOutputTokens 5-10× higher than your visible-output budget, or use @llm-ports/adapter-openai directly for reasoning models in v0.1. Tracked at #4.
• generateStructured throws SyntaxError on empty model responses. When a reasoning model returns an empty completion (above), the JSON parser throws SyntaxError: Unexpected end of JSON input, which currently wraps as a generic ProviderUnavailableError. v0.2 ships a more specific EmptyResponseError class. Tracked at #5.

Reading next

• Migration from Vercel AI SDK →
• Adapter feature matrix → — when to use this vs direct adapters