Grok3API

An asynchronous library for interacting with Grok via the mobile gRPC channel: sending messages, streaming responses, parsing protobuf chunks, and retrieving structured API data.

---

➡ Ru README

Features

• asynchronous client
• streaming responses via new_ask_stream()
• regular requests via new_ask()
• protobuf/gRPC framing
• typed Pydantic models
• streaming chunk parsing
• REST/gRPC/stream error handling
• support for metadata/tool responses/search results

Installation

pip install git+ssh://git@github.com/boykopovar/Grok3API.git -U

Requirements

• Python 3.8+
• aiohttp
• coincurve

Streaming Example

import asyncio

from grok3api.client import GrokClient
from grok3api.types.request import ChatRequest
from grok3api.types import TokenChunk


async def main():
    async with GrokClient() as client:
        while True:
            print("\nGrok: ", end="")

            async for chunk in client.new_ask_stream(
                request=ChatRequest(
                    message=input("\nYou: "),
                    temporary=False
                ),
                chunks_white_list=(TokenChunk,)
            ):
                print(chunk.token, end="", flush=True)


if __name__ == "__main__":
    asyncio.run(main())

Standard Request

import asyncio

from grok3api.client import GrokClient
from grok3api.types.request import ChatRequest


async def main():
    async with GrokClient() as client:
        response = await client.new_ask(
            ChatRequest(
                message="Hello",
                temporary=False
            )
        )

        print(response.text)
        print(response.model_response)


if __name__ == "__main__":
    asyncio.run(main())

Continue started conversation

import asyncio

from grok3api.client import GrokClient
from grok3api.types.request import ChatRequest, AddResponseRequest


async def main():
    async with GrokClient() as client:
        first = await client.new_ask(
            ChatRequest(
                message="Hello",
                temporary=False
            )
        )

        second = await client.add_response(
            AddResponseRequest(
                conversation_id=first.conversation.conversation_id,
                message="How are you?"
            )
        )

        print(second.text)


if __name__ == "__main__":
    asyncio.run(main())

Pydantic Models

All models fully match the actual protobuf/gRPC API responses.

The library does not include:

• simplified structures
• artificial abstraction layers
• field normalization
• payload field renaming
• hiding original API values

Supported

Response Chunks

• TokenChunk
• ModelResponseChunk
• FinalMetadataChunk
• SurveyChunk
• ConversationChunk
• TitleChunk

Response Models

• AskResponse
• ModelResponse
• FinalMetadata
• Conversation
• Survey
• ResponseStep
• ToolUsageResult
• WebSearchResult
• XPost
• RagResult
• CollectionSearchResultItem
• ConnectorSearchResultItem
• StreamError

Tool Responses

• web search
• X/Twitter search
• RAG results
• connector search
• collection search
• memory updates
• code execution
• image generation
• video generation
• audio generation

Error Handling

from grok3api.types.exceptions import (
    GrokApiError,
    GrokRestError,
    GrokGrpcError,
    GrokRateLimitError,
    GrokUnderHeavyUsageError,
    GrokStreamError,
    GrokUnavailableRegionError,
    GrokTooManyRequestsError
)

How the Client Works

1. the protobuf request is serialized into a binary payload
2. the payload is wrapped into a gRPC frame
3. the request is sent to the mobile gRPC endpoint
4. the client reads the streaming response
5. protobuf chunks are decoded into Pydantic models

Disclaimer

This project is not affiliated with xAI and is not an official SDK.

This is an independent implementation of the mobile Grok API obtained through reverse engineering.