Yandex AI Studio SDK

The yandex-cloud-ml-sdk package is a backward compatibility wrapper for the yandex-ai-studio-sdk. If you are maintaining legacy codebases that depend on the older naming conventions or structures, you can use this package.

For detailed instructions on how to transition from the legacy SDK to the current version, refer to the migration guide.

The Yandex AI Studio SDK provides several modules to define how search indexes are constructed, including file-based indexing, chunking strategies, combination strategies, and normalization strategies.

Key components include:

• SearchIndexFile: Used for creating indexes from local files.
• index_type: Defines the fundamental type of the search index.
• chunking_strategy: Determines how text is split into smaller segments (chunks) for indexing.
• combination_strategy: Defines how chunks or results are combined.
• normalization_strategy: Defines how data is normalized before indexing or retrieval.

The AsyncSpeechKitDomain class is the entry point for interacting with Yandex SpeechKit services asynchronously within the SDK. It provides access to specialized domains for speech processing, specifically:

• Text-to-Speech: Converting text into synthesized speech.
• Speech-to-Text: Transcribing spoken audio into text.

To use these features, you should access the sub-domains provided by an instance of AsyncSpeechKitDomain.

from yandex_ai_studio_sdk._speechkit.domain import AsyncSpeechKitDomain

# An instance of AsyncSpeechKitDomain provides access to:
# - domain.text_to_speech
# - domain.speech_to_text

The SDK provides Vision capabilities, specifically Optical Character Recognition (OCR), to extract text from images. The OCR implementation is structured around three main components:

1. OCRConfig: Used to configure the parameters of the OCR request.
2. OCRResult: The object returned containing the extracted text and metadata.
3. yandex_ai_studio_sdk._vision.ocr.text_annotation: Provides the underlying text annotation structures used to describe the location and content of recognized text.

To perform OCR, you will typically instantiate an OCRConfig object and pass it to the vision service methods to receive an OCRResult.

The yandex-cloud-ml-sdk package serves as a compatibility wrapper for yandex-ai-studio-sdk starting from version 0.19.0.

• The last standalone version of yandex-cloud-ml-sdk is 0.18.
• Version 0.19.0 of yandex-cloud-ml-sdk is a wrapper that points to yandex-ai-studio-sdk version 0.19.0.

If you are using yandex-cloud-ml-sdk version 0.19.0 or higher, you are effectively using the yandex-ai-studio-sdk via a compatibility layer.

If you intend to use an agent with a code interpreter, you must perform these additional configuration steps in Open WebUI to ensure files are correctly passed to the agent rather than being handled by Open WebUI's internal RAG:

1. Model Selection: Locate the model corresponding to your selected agent in the Open WebUI admin interface at /admin/settings/models.
2. Disable File Context: In the model settings, turn off File Context. This prevents Open WebUI from attaching RAG context to user messages, allowing this pipe to attach the files directly to your request for code interpretation.
3. Optimize File Attachment (Optional): To increase the speed of attaching files to the WebUI, navigate to /admin/settings/documents and enable Bypass Embedding and Retrieval.

The SDK includes a CLI for building vector search indexes from various sources. Depending on your data source, you may need to install specific extras:

• Base CLI (Local files + Confluence): pip install yandex-ai-studio-sdk
• S3 Support (S3 / Yandex Object Storage): pip install "yandex-ai-studio-sdk[cli-s3]"
• Wiki Support (Wikipedia / MediaWiki): pip install "yandex-ai-studio-sdk[cli-wiki]"

# Install base CLI (local files + Confluence)
pip install yandex-ai-studio-sdk

# With S3 / Yandex Object Storage support
pip install "yandex-ai-studio-sdk[cli-s3]"

# With Wikipedia / MediaWiki support
pip install "yandex-ai-studio-sdk[cli-wiki]"

The CLI resolves authentication automatically in the following order:

1. Explicit credentials via --auth, YC_API_KEY, or YC_IAM_TOKEN.
2. Yandex Cloud CLI (yc) if installed and configured.
3. Compute Metadata Service (automatic on Yandex Compute Cloud VMs).

Note: YC_API_KEY takes precedence over YC_IAM_TOKEN if both are set.

Required Environment Variables:

• YC_FOLDER_ID: Your Yandex Cloud folder ID.

Optional Authentication Variables:

• YC_API_KEY: Your API key.
• YC_IAM_TOKEN: Your IAM token.

export YC_FOLDER_ID=b1gxxxxxxxxxxxxx
export YC_API_KEY=AQVNxxxxxxxxxxxxxx
# or
export YC_IAM_TOKEN=t1.9euelZxxxxxxxxxxxxxx

To use the SDK with LangChain, you must install the package with the langchain extra. Once installed, you can convert a completion model into a LangChain-compatible object using the .langchain() method. This allows you to use standard LangChain methods like .invoke() with HumanMessage and AIMessage objects.

pip install yandex-ai-studio-sdk[langchain]

from yandex_ai_studio_sdk import AIStudio
from langchain_core.messages import AIMessage, HumanMessage

sdk = AIStudio(folder_id="...", auth="<APIKey/IAMToken/SomethingElse>")

model = sdk.models.completions('yandexgpt').langchain()

langchain_result = model.invoke([
    HumanMessage(content="hello!"),
    AIMessage(content="Hi there human!"),
    HumanMessage(content="Meow!"),
])

This integration allows you to use agents created in the Yandex AI Studio service as manifold pipes within Open WebUI.

Setup Steps

1. Create an Agent: Visit https://aistudio.yandex.ru/ and create your desired agent.
2. Obtain API Key: Retrieve your API Key from https://aistudio.yandex.ru/platform/folders/<your_folder_id>/access.
3. Import Pipe: Import this manifold pipe into your Open WebUI instance.
4. Configure Valves: In Open WebUI, configure the following valve settings:
   • Yandex Cloud Api Key: The key obtained in step 2.
   • Yandex Cloud Folder Id: Your Yandex Cloud folder ID.
   • Agent Ids: The IDs of the agents created in step 1 (supports one or more IDs separated by commas).
   • Agent Names (Optional): Custom names for the agents to appear more cleanly in the WebUI.