Finder Tools#

Configuration File: finder_tools.json Tool Type: Local Tools Count: 5

This page contains all tools defined in the finder_tools.json configuration file.

Available Tools#

Tool_Finder (Type: ToolFinderEmbedding)#

Retrieve related tools from the toolbox based on the provided description, advanced version with …

Tool_Finder tool specification

Tool Information:

  • Name: Tool_Finder

  • Type: ToolFinderEmbedding

  • Description: Retrieve related tools from the toolbox based on the provided description, advanced version with more functionality.

Parameters:

  • description (string) (required) The description of the tool capability required.

  • limit (integer) (optional) The number of tools to retrieve (default: 10)

  • picked_tool_names (array) (optional) Pre-selected tool names to process. If provided, tool selection will skip these tools.

  • return_call_result (boolean) (optional) Whether to return both prompts and tool names. If false, returns only tool prompts.

  • categories (array) (optional) Optional list of tool categories to filter by

  • embedding_model (string) (optional) Optional: embedding encoder to use for this search. ‘default’ = the fine-tuned ToolRAG-T1 (1.5B); ‘gte-qwen2-7b’ and ‘e5-mistral-7b’ are larger open encoders with higher accuracy (used only if already in the local model cache unless TOOLUNIVERSE_ALLOW_ENCODER_DOWNLOAD=1; GPU recommended); ‘openai-3-large’/’openai-3-small’ use hosted OpenAI embeddings (require OpenAI/Azure credentials). Defaults to ‘default’.

Example Usage:

query = {
    "name": "Tool_Finder",
    "arguments": {
        "description": "example_value"
    }
}
result = tu.run(query)

Tool_Finder_Jev (Type: ToolFinderJev)#

Tool finder that retrieves a wide BM25 shortlist and grades every candidate with a hosted decisio…

Tool_Finder_Jev tool specification

Tool Information:

  • Name: Tool_Finder_Jev

  • Type: ToolFinderJev

  • Description: Tool finder that retrieves a wide BM25 shortlist and grades every candidate with a hosted decision model, for higher accuracy than embedding search without a GPU. Requires TYPESAFE_API_KEY. Grading sends the search text and the descriptions of the retrieved tools to TypeSafe’s hosted service at https://api.typesafe.ai/v1/systemone, so do not submit sensitive or patient-identifying information. If grading is unavailable the BM25 retrieval order is returned instead, flagged with graded=false.

Parameters:

  • description (string) (required) The description of the tool capability required.

  • limit (integer) (optional) The number of tools to retrieve (default: 10)

  • picked_tool_names (array) (optional) Pre-selected tool names to process. If provided, retrieval and grading are skipped.

  • return_call_result (boolean) (optional) Whether to return both prompts and tool names. If false, returns only tool prompts.

  • categories (array) (optional) Optional list of tool categories to filter by

Example Usage:

query = {
    "name": "Tool_Finder_Jev",
    "arguments": {
        "description": "example_value"
    }
}
result = tu.run(query)

Tool_Finder_Keyword (Type: ToolFinderKeyword)#

Simple keyword-based tool finder for discovering relevant tools using text matching

Tool_Finder_Keyword tool specification

Tool Information:

  • Name: Tool_Finder_Keyword

  • Type: ToolFinderKeyword

  • Description: Simple keyword-based tool finder for discovering relevant tools using text matching

Parameters:

  • description (string) (required) The description of the tool capability required.

  • limit (integer) (optional) The number of tools to retrieve (default: 10)

  • picked_tool_names (array) (optional) Pre-selected tool names to process. If provided, tool selection will skip these tools.

  • return_call_result (boolean) (optional) Whether to return both prompts and tool names. If false, returns only tool prompts.

  • categories (array) (optional) Optional list of tool categories to filter by

Example Usage:

query = {
    "name": "Tool_Finder_Keyword",
    "arguments": {
        "description": "example_value"
    }
}
result = tu.run(query)

Tool_Finder_LLM (Type: ToolFinderLLM)#

LLM-based tool finder that uses natural language processing to intelligently select relevant tool…

Tool_Finder_LLM tool specification

Tool Information:

  • Name: Tool_Finder_LLM

  • Type: ToolFinderLLM

  • Description: LLM-based tool finder that uses natural language processing to intelligently select relevant tools based on user queries. This tool analyzes all available tool descriptions and uses an LLM to determine which tools would be most helpful for a given task or question.

Parameters:

  • description (string) (required) The description of the tool capability required.

  • limit (integer) (optional) The number of tools to retrieve (default: 10)

  • picked_tool_names (array) (optional) Pre-selected tool names to process. If provided, tool selection will skip these tools.

  • return_call_result (boolean) (optional) Whether to return both prompts and tool names. If false, returns only tool prompts.

  • categories (array) (optional) Optional list of tool categories to filter by

Example Usage:

query = {
    "name": "Tool_Finder_LLM",
    "arguments": {
        "description": "example_value"
    }
}
result = tu.run(query)

Tool_RAG (Type: ToolFinderEmbedding)#

Retrieve related tools from the toolbox based on the provided description

Tool_RAG tool specification

Tool Information:

  • Name: Tool_RAG

  • Type: ToolFinderEmbedding

  • Description: Retrieve related tools from the toolbox based on the provided description

Parameters:

  • description (string) (required) The description of the tool capability required.

  • limit (integer) (optional) The number of tools to retrieve (default: 10)

  • embedding_model (string) (optional) Optional: embedding encoder to use for this search. ‘default’ = the fine-tuned ToolRAG-T1 (1.5B); ‘gte-qwen2-7b’ and ‘e5-mistral-7b’ are larger open encoders with higher accuracy (used only if already in the local model cache unless TOOLUNIVERSE_ALLOW_ENCODER_DOWNLOAD=1; GPU recommended); ‘openai-3-large’/’openai-3-small’ use hosted OpenAI embeddings (require OpenAI/Azure credentials). Defaults to ‘default’.

Example Usage:

query = {
    "name": "Tool_RAG",
    "arguments": {
        "description": "example_value"
    }
}
result = tu.run(query)