Detect if there are any URLs in the messages. Args: messages: List of message dictionaries

Generate embeddings for an image using the ollama model. Theoretically, OllamaClient.embed could directly be called from self.genera

Generate embeddings using native Cohere SDK. Args: texts: List of texts to embed model: Embedding mo

Generate embeddings using OpenAI compatibility API. Args: texts: List of texts to embed model: Embed

Generate embeddings for an image using the ollama model. Theoretically, OllamaClient.embed could directly be called from self.genera

Generate embeddings for a video using the ollama model. Currently, an embedding of a video is represented by a set of embeddings for

Converts ChromaDB query result to Embedding Object.

Converts Qdrant query result to Embedding Object.

Enforce task limit by evicting oldest tasks if necessary.

Ensure the specified model is available on the Lemonade server.

Robust method to ensure a new TextEdit document is created. :return: Tuple of (success, message)

Escape newline characters within JSON string values.

Escape newline characters within JSON string values.

Escape newlines in JSON strings for better parsing.

Evaluate a single minions run based on the specified metric Args: result: Result dictionary from minions run

Execute the appropriate action based on the action type in the JSON. WARNING: This method executes actions on the host machine based

Execute a single development step with local LLM and remote review.

Execute Minions (parallel) query.

Execute a specific Minions skill. Expected metadata format for Minions integration: - skill_id: "minion_query" or "minions_qu

Execute tool calls and return the results. Args: tool_calls: List of tool call objects from the LLM custom_tool_execu

Extract a complete JSON object by counting balanced braces.

Extract citations from response if available.

Extract and load the first image referenced in messages. Supports keys: 'images' (list/str), 'image' (str), 'image_url' (data URL or

Extract text content from an Interactions API response. Args: response: The Interactions API response object

Extract usage information from an Interactions API response. Args: response: The Interactions API response objec

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

Extract text content from a PDF file

Extract token usage from Agno response if available.

Find the index of the first user message in the messages list.

Format completed steps summary with test results.

Format the conversation history for inclusion in prompts.

Format messages for the Interactions API. Args: messages: List of message dictionaries with 'role' and 'content'

Format a message with multimodal content for the Qwen2.5-Omni model. Args: message: Message dictionary with 'role' and '

Format a message with multimodal content for the Qwen2.5-Omni model. Args: message: Message dictionary with 'role' and '

Format namespace variables for JSON logging.

Generate text without streaming.

Generate text without streaming.

Generate markdown using MLX implementation.

Generate markdown using Transformers implementation.

Generate explanation of available MCP tools for supervisor.

Generate the development runbook using the remote LLM.

Determine the appropriate device and dtype to use. Returns: tuple: (device, dtype)

Get Tencent-specific extra_body parameters for API requests. Returns: Dictionary of extra_body parameters for Hunyuan API

Get OpenRouter-specific headers.

Handle tasks/cancel requests.

Handle tasks/get requests.

Handle tasks/send requests.

Handle tasks/sendSubscribe requests with streaming.

Handle streaming response and collect full content.

Import Minions modules and clients.

Initialize the client in a separate thread

Initialize the appropriate client based on configuration.

Initialize the appropriate client based on configuration.

Initialize vLLM with LMCache integration.

Set up the secure communication channel with attestation and key exchange

Detect ServiceNow Apriel models which use AutoModelForImageTextToText.

Detect Apple FastVLM models which require special image handling.

Check if the current model is a Hunyuan model. Returns: bool: True if the model is a Hunyuan model, False otherw

Check if the system is a Mac.

Detect Facebook MobileLLM models which use subfolders for different versions.

Check if the model is a Nemotron model. Args: model_name: The model name to check Returns:

Check if nvidia-smi is available on the system.

Check if the given model supports reasoning. Args: model_name: The model name to check Returns: boo

Check if the current session is still valid

Last-chance attempt: download the JWKS and try every key until one verifies the signature. Works even when there's no `kid` or `x5c`.

Load API keys from file.

Load OAuth2 clients from storage.

Load MCP configuration from file

Load default configuration from environment variables.

Load a separate model for generating embeddings.

Have local LLM implement a development step.

Log the story generation session.

Make a chat request to the Docker Model Runner API

Make an async chat request to the Docker Model Runner API

Convert chat messages to a single prompt string for the Agno Agent. The Agno Agent expects a single prompt, so we need to co

Convert chat messages to a single prompt string. This is a simple implementation - you may want to use proper chat templates

Create a session header by hashing all slice metadata.

Verify and decrypt a single chunk of data.

Parse Apriel model response to extract thinking content and final answer. Args: completion_text: Raw completion

Parse Hunyuan model response to extract thinking content and answer content. Args: completion_text: Raw completi

Prepare kwargs for Nemotron models, including reasoning budget control. Args: **kwargs: Original kwargs

Prepare messages for Nemotron models by adding appropriate reasoning control. Args: messages: Original messages

Prepare the structured output schema. Args: schema: Pydantic model class or JSON Schema dict

Process a folder containing text, PDF, and image files.

Process various image input types and return a PIL Image.

Process a media file (image, audio, video) from a file path or URL. Args: file_path_or_url: Path to local file or URL

Process a media file (image, audio, video) from a file path or URL. Args: file_path_or_url: Path to local file or URL

Pull the model using Docker Model Runner.

Raise an exception with Docker installation instructions.

Have remote LLM review the implementation.

Run the task with per-turn decisions between local and remote models while following minion protocol.

Run the entire task using only the local model.

Run the entire task using only the remote model.