streamUI with anthropic models

[amirm363]

In the documentation it says that: "Anthropic language models can also be used in the streamText, generateObject, streamObject, and streamUI functions (see AI SDK Core and AI SDK RSC)."

until now I've worked only with OpenAI models with the streamUI function and everything worked flawlessly, I want to try anthropic models now, but it looks like the "model" property in the streamUI options can't use the anthropic provider (even though the docs says it does), this is the error: "Type 'AnthropicMessagesLanguageModel' is not assignable to type 'LanguageModelV1'."

const result = await streamUI({
system,
model: anthropic('claude-3-haiku-20240307'),
maxTokens: 1024,
messages: [...history.get()],
text: ({ content, done }) => {...},
tools: {...}
temperature: 0,
});
return {... };
}

what am I missing here?

[lgrammel]

Have you updated @ai-sdk/anthropic and ai to their latest versions? It could be a version mismatch (older version of ai-sdk/anthropic).

[amirm363]

Thats exactly what I've done, reran the project and now it works, the problem now that tool calls that worked perfectly fine with openAI sends an Error object to the client, which crushes the application.

Warning: Only plain objects can be passed to Client Components from Server Components. Error objects are not supported.
{error: Error}
^^^^^
Warning: Only plain objects can be passed to Client Components from Server Components. Error objects are not supported.
{name: ..., message: ..., cause: Error, stack: ..., toolName: ..., toolArgs: ...}
^^^^^
⨯ Error: Only plain objects, and a few built-ins, can be passed to Client Components from Server Components. Classes or null prototypes are not supported.
at stringify ()

is there anyway tracing it?

[retired-Hurt]

I am having the same issue on all of the non openai providers like Google or anthropic or mistral. The vscode compiler gives an error saying below error. I've reinstalled the latest versions just to double check. I suspect this might be something at a framework level and not specific to a particular provider.

Type 'GoogleVertexLanguageModel' is not assignable to type 'LanguageModelV1'.
The types of 'doGenerate(...).then' are incompatible between these types.
Type '<TResult1 = { text?: string | undefined; toolCalls?: import("/Users/code/nextjs-chat-play/node_modules/.pnpm/@AI-SDK[email protected]/node_modules/@ai-sdk/provider/dist/index").LanguageModelV1FunctionToolCall[] | undefined; ... 5 more ...; logprobs?: import("/Users/code/nextjs-...' is not assignable to type '<TResult1 = { text?: string | undefined; toolCalls?: import("/Users/code/nextjs-chat-play/node_modules/.pnpm/@AI-SDK[email protected]/node_modules/@ai-sdk/provider/dist/index").LanguageModelV1FunctionToolCall[] | undefined; ... 5 more ...; logprobs?: import("/Users/code/nextjs-c...'.
Types of parameters 'onfulfilled' and 'onfulfilled' are incompatible.
Types of parameters 'value' and 'value' are incompatible.
Type '{ text?: string | undefined; toolCalls?: import("/Users/code/nextjs-chat-play/node_modules/.pnpm/@AI-SDK[email protected]/node_modules/@ai-sdk/provider/dist/index").LanguageModelV1FunctionToolCall[] | undefined; ... 5 more ...; logprobs?: import("/Users/code/nextjs-chat-play/no...' is not assignable to type '{ text?: string | undefined; toolCalls?: import("/Users/code/nextjs-chat-play/node_modules/.pnpm/@AI-SDK[email protected]/node_modules/@ai-sdk/provider/dist/index").LanguageModelV1FunctionToolCall[] | undefined; ... 5 more ...; logprobs?: import("/Users/code/nextjs-chat-play/nod...'.
Types of property 'finishReason' are incompatible.
Type 'import("/Users/code/nextjs-chat-play/node_modules/.pnpm/@AI-SDK[email protected]/node_modules/@ai-sdk/provider/dist/index").LanguageModelV1FinishReason' is not assignable to type 'import("/Users/code/nextjs-chat-play/node_modules/.pnpm/@AI-SDK[email protected]/node_modules/@ai-sdk/provider/dist/index").LanguageModelV1FinishReason'.
Type '"unknown"' is not assignable to type 'LanguageModelV1FinishReason'.ts(2322)
index.d.ts(467, 5): The expected type comes from property 'model' which is declared here on type 'CallSettings & Prompt & { model: LanguageModelV1; tools?: {} | undefined; text?: RenderText | undefined; initial?: ReactNode; }'

[lgrammel]

There is a version mismatch there. One of your dependencies is old and uses @ai-sdk/provider v0.0.3, another one is newer and uses v0.0.10

Are you sure you are on the latest versions? the latest version of e.g. the anthropic provider needs v0.0.10: https://github.com/vercel/ai/blob/main/packages/anthropic/package.json

[lionstone]

i believe i am having the same issue - streamObject works fine with openai but not with anthropic

[lgrammel]

@lionstone have you updated the anthropic provider and the ai sdk to the latest version? which error are you seeing?