vercel/ai · error · UnsupportedFunctionalityError

The ${model.provider} model "${model.modelId}" does not supp

Error message

The ${model.provider} model "${model.modelId}" does not support batch processing.

What it means

The ACP harness fingerprints the active host tool catalog (SHA-256 of the JSON-serialized HarnessV1ToolSpec array) and persists it with the turn. During a lossy rerun (a resumed turn whose full protocol state no longer exists), assertRecoveryToolCatalog compares the persisted fingerprint against the currently configured tools. A mismatch means the rerun would execute against a different tool surface than the original turn, so the harness refuses to proceed rather than replaying with wrong tools.

Source

Thrown at packages/ai/src/batch/batch.ts:224

      await stream.pipeTo(transform.writable, {
        signal: operationAbortSignal,
      });
    } catch (error) {
      await transform.writable.abort(wrapGatewayError(error)).catch(() => {});
    }
  })();

  return asAsyncIterableStream(transform.readable);
}

function resolveBatchLanguageModel(
  modelArg: StartTextBatchOptions['model'],
): BatchLanguageModelV4 {
  const model = resolveLanguageModel(modelArg);

  if (!isBatchLanguageModel(model)) {
    throw new UnsupportedFunctionalityError({
      functionality: 'batch processing',
      message: `The ${model.provider} model "${model.modelId}" does not support batch processing.`,
    });
  }

  return model;
}

function isBatchLanguageModel(
  model: LanguageModelV4,
): model is BatchLanguageModelV4 {
  const candidate = model as Partial<BatchLanguageModelV4>;
  return (
    typeof candidate.experimental_doStartBatch === 'function' &&
    typeof candidate.experimental_doGetBatchStatus === 'function' &&
    typeof candidate.experimental_doGetBatchResults === 'function'
  );
}

View on GitHub (pinned to 69428b1f8b)

Solutions

1. Restore the exact `tools` array (names, descriptions, parameter schemas, order) that was active when the original turn ran.
2. If the tool change is intentional, start a new session/turn instead of rerunning the old one.
3. Diff the persisted tool catalog from your lifecycle/turn state against the current HarnessV1ToolSpec[] to identify what changed.
4. Pin tool definitions (including MCP server versions) so they are stable across deploys for the lifetime of a session.

Example fix

// before
classifierAgent({ tools: [readFileTool, writeFileTool, grepTool] }) // original turn
// rerun after removing grepTool
agent({ tools: [readFileTool, writeFileTool] }) // throws

// after
agent({ tools: [readFileTool, writeFileTool, grepTool] }) // identical catalog

import { createHash } from 'node:crypto';
const fingerprint = (tools: unknown) =>
  createHash('sha256').update(JSON.stringify(tools)).digest('hex');

// before rerunning, compare against the fingerprint persisted with the turn
if (fingerprint(currentTools) !== persistedToolCatalogFingerprint) {
  throw new Error('Tool catalog changed since the original turn; start a new session.');
}

try {
  await agent.rerunTurn({ turnId });
} catch (error) {
  if (error instanceof Error && error.message.includes('same active host tool catalog')) {
    // tool catalog drifted: recreate the session with the current catalog
    session = await createSession({ tools: currentTools });
  } else {
    throw error;
  }
}

When it happens

Trigger: Resuming/rerunning a previous ACP turn after the `tools` array passed to the harness (or the resolved builtin + MCP tool catalog) changed between the original turn and the rerun; adding, removing, renaming, or reordering host tools; changing tool parameter schemas; switching mcpServers so the effective catalog differs.

Common situations: Deploying a new app version that added or removed a tool while a session was in-flight; toggling MCP servers via env-driven config; a tool library upgrade that changed a tool's name or parameters; reordering tools and assuming order is irrelevant.

Related errors

• ${harnessId} recovered this turn through disk replay only an
• The ACP implementation did not load the active harness-owned
• Invalid argument for parameter requests: requests must not b
• Invalid argument for parameter requests: request IDs must no
• Invalid argument for parameter requests: request IDs must be

AI-assisted analysis of vercel/ai@69428b1f8b (2026-08-30). Data as JSON: /api/errors/dcffbef76685c0df. Report an issue: GitHub.