vercel/ai · error

Unknown chunk type: ${exhaustiveCheck}

Error message

Unknown chunk type: ${exhaustiveCheck}

What it means

A fallback `never`-exhaustiveness guard in streamText's internal chunk-type switch. It fires only when the runtime delivers a chunk type the installed SDK version doesn't know, indicating a version mismatch between the ai package and a provider package (or internal state corruption).

Source

Thrown at packages/ai/src/generate-text/stream-text.ts:2488

                    }

                    case 'error': {
                      hasReceivedTerminalChunk = true;
                      controller.enqueue(chunk);
                      stepFinishReason = 'error';
                      break;
                    }

                    case 'raw': {
                      if (include.rawChunks) {
                        controller.enqueue(chunk);
                      }
                      break;
                    }

                    default: {
                      const exhaustiveCheck: never = chunkType;
                      throw new Error(`Unknown chunk type: ${exhaustiveCheck}`);
                    }
                  }
                },

                // invoke onEnd callback and resolve toolResults promise when the stream is about to close:
                async flush(controller) {
                  // emit an error when an incomplete model stream produced no
                  // output instead of recording an empty step. incomplete
                  // streams with partial output retain the partial result:
                  if (!hasReceivedTerminalChunk && !hasReceivedOutputChunk) {
                    controller.enqueue({
                      type: 'error',
                      error: new NoOutputGeneratedError({
                        message:
                          'No output generated. The model stream ended without a finish chunk.',
                      }),
                    });

View on GitHub (pinned to 69428b1f8b)

Solutions

  1. Run `pnpm why @ai-sdk/provider` (or npm ls) and align all @ai-sdk packages and `ai` to versions from the same release.
  2. Update `ai` to the latest version so new provider part types are handled.
  3. Check any stream middleware/custom transforms for emitting chunk types outside LanguageModelV* streams.
  4. Reinstall node_modules to clear duplicated transitive provider packages.

Example fix

// before
"dependencies": { "ai": "^4.0.0", "@ai-sdk/openai": "^2.0.0" } // mismatched majors
// after
"dependencies": { "ai": "^4.3.16", "@ai-sdk/openai": "^1.3.0" } // matching release lines
Defensive patterns

Strategy: fallback

Validate before calling

// before starting, verify a single provider version is installed:
// pnpm why @ai-sdk/provider  -> expect exactly one version in the tree
import { VERSIONS } from '@ai-sdk/provider-utils'; // log during support debugging
console.log('provider-utils version:', VERSIONS);

Try / catch

try {
  await streamText({ model, prompt });
} catch (e) {
  if (e instanceof Error && e.message.startsWith('Unknown chunk type')) {
    console.error('Incompatible ai/@ai-sdk provider versions; upgrade `ai`.');
    // fall back to non-streaming generateText or a raw provider client
  } else throw e;
}

Prevention

When it happens

Trigger: Mixing incompatible versions of `ai` and `@ai-sdk/<provider>` (e.g. provider emits a new part type the older ai core doesn't handle); patching/monkey-patching stream parts; custom middleware emitting an unknown chunk type.

Common situations: Partial pnpm dependency resolution leaving two major versions of @ai-sdk/provider in node_modules; upgrading a provider package without upgrading `ai`; third-party middleware producing nonstandard parts.

Related errors

AI-assisted analysis of vercel/ai@69428b1f8b (2026-08-30). Data as JSON: /api/errors/cbdab4272dc78e86. Report an issue: GitHub.