Handle chat completions using Exa's OpenAI-compatible API. This routes to Exa's /chat/completions endpoint which provides

Assert response contains expected content

Log test result.

Run the server.

Encode texts to create embeddings using MLX model. Args: texts: Single text or list of texts to encode model

Log test result.

Log test result.

Assert that chat response has correct format

Close the test client.

Generate current metrics in Prometheus format.

Standard test messages for consistency

Create a new task.

Search the web using Exa's native search API. Supports multiple search types including instant search for sub-150ms latency.

BM25 retrieval using keywords and weights.

Send JSON-RPC request with authentication.

Track authentication attempt.

Extract query and document from A2A message parts. Args: parts: List of A2A parts (text, file, or data) Returns:

Find all unique matches for a regex pattern in the text.

Check if file is an image.

Check if token has required scopes.

Generate embeddings for the given content. Args: content: Text content to embed (single string or list of string

Generate a new API key.

Get or initialize the MLX embedding model and tokenizer.

Register a new OAuth2 client.

Start background tasks. Must be called when event loop is running.

Extract text content from file information.

Extract JSON from text that may be wrapped in markdown code blocks.

Generate a unique key for client configuration.

Make HTTP request with error handling.

Streamlit renamed `experimental_rerun` → `rerun` in v1.27. Call the one that exists.

Retrieves top k chunks using dense vector embeddings and FAISS similarity search Args: queries: List of query strings chunks

Extract query and document from A2A parts following the new format. Expected: First part = query (text), Second part = document (text

Route a query to the most appropriate client using Arch-Router's XML format. Args: query: User query to route

Send a task to the A2A server.

Extract JSON from text that may be wrapped in markdown code blocks.

Start MCP server (e.g. 'github' or 'filesystem') and make a Minion with access to it

Handle synchronous chat completions. If you pass a list of message dicts, we do one call for that entire conversation. If you pass a

Clear all turns from the history.

Format parameter count in human-readable form.

Get a list of available models from Exa. Returns: List[str]: List of model names available through Exa API

Get current state of the workspace. Returns: Dictionary containing file structure and contents

Get task status and results.

Send a task to the A2A server.

Stop the background monitoring thread.

Update task status.

Validate an API key.

Wait for a task to complete with progress logging.

Ensure the native xAI SDK is initialized. Raises: ImportError: If xai_sdk is not installed

Handle chat completions using the OpenAI API. Args: messages: List of message dictionaries with 'role' and 'content' key

Handle chat completions using the Moonshot AI API. Args: messages: List of message dictionaries with 'role' and 'content

Create the appropriate Minions protocol instance.

Execute a task.

Set up the secure communication channel with attestation and key exchange

Parse A2A task metadata to create Minions configuration.

Retrieve entries from the Qdrant collection based on query. :param query: embedding object (TextEmbedding) of query :param t

Start the background monitoring thread.

Track a streaming event.

Validate OAuth2 client credentials.

Verify a JWT token.

Handle async chat with multiple messages in parallel.

Handle async chat with multiple messages in parallel. Args: messages: List of message dictionaries with 'role' and 'cont

Handle async chat with MCP tool calling support.

Extract JSON from text that may be wrapped in markdown code blocks.

Extract JSON from text that may be wrapped in markdown code blocks.

Clean common LLM JSON issues and parse.

Extract text from PDF bytes asynchronously.

Check if the current model is a reasoning model (GPT-5 family). Returns: True if the model supports reasoning ef

Save file content to temporary file and return path.

Convert usage objects to JSON-serializable format.

Handle chat completions using the Qwen API. Args: messages: List of message dictionaries with 'role' and 'content' keys

Handle chat completions using the HuggingFace Inference API or Router. Args: messages: List of message dictionaries with

Clean up LMCache resources.

Create a JWT token from token data.

Encode texts to create ColBERT embeddings. Args: texts: Single text or list of texts to encode is_qu

Extract skill ID from metadata or raise error if missing.

Generate audio from text using CSM-MLX. Args: text (str): The text to convert to speech voice (str): Voice s

Create HTML for audio playback from base64 encoded audio. Args: audio_base64 (str): Base64 encoded audio data R