continuedev/continue · error · Error

AI SDK provider does not support reranking.

Error message

AI SDK provider does not support reranking.

What it means

The AiSdk adapter implements rerank as a throwing stub because the Vercel AI SDK has no reranking primitive. Any rerank request routed to an AI SDK provider fails immediately. Use an adapter backed by a rerank-capable service (Cohere, Jina, Mistral).

Source

Thrown at packages/openai-adapters/src/apis/AiSdk.ts:309

    }

    const result = await embedMany({
      model,
      values: stringInputs,
    });

    return embedding({
      data: result.embeddings,
      model: modelId,
      usage: {
        prompt_tokens: result.usage?.tokens ?? 0,
        total_tokens: result.usage?.tokens ?? 0,
      },
    });
  }

  async rerank(_body: RerankCreateParams): Promise<CreateRerankResponse> {
    throw new Error("AI SDK provider does not support reranking.");
  }

  async list(): Promise<Model[]> {
    return [];
  }
}

View on GitHub (pinned to 5522c6f44c)

Solutions

  1. Use a rerank-capable adapter (CohereRerankAdapter, Jina, Mistral) for the rerank step
  2. Skip reranking when the adapter doesn't support it, falling back to pure embedding similarity
  3. Check adapter capabilities before building the retrieval pipeline

Example fix

// before
const res = await api.rerank({ query, documents, model: 'rerank-v1' });

// after
const res = await cohereApi.rerank({ query, documents, model: 'rerank-v3.5' });
Defensive patterns

Strategy: fallback

Validate before calling

if (api instanceof AiSdk) { return rankByEmbedding(query, docs); }

Type guard

function supportsRerank(api: BaseApi): boolean { return !(api instanceof AiSdk); }

Try / catch

try { return await api.rerank(body); }
catch (e) { if (/does not support reranking/.test(String(e))) return fallbackEmbeddingRerank(body); throw e; }

Prevention

When it happens

Trigger: Calling rerank(body) on an AiSdk adapter instance, typically from retrieval/RAG pipelines that rerank documents.

Common situations: Building RAG with an AI SDK provider (e.g. an AI SDK-hosted model) and assuming rerank works uniformly across all adapters because the interface exposes it.

Related errors


AI-assisted analysis of continuedev/continue@5522c6f44c (2026-08-27). Data as JSON: /api/errors/0debd9e8634fbbd8. Report an issue: GitHub.