LLM Tool

Usage

Portia offers both open source tools as well as a cloud-hosted library of tools to save you development time. You can dig into the specs of those tools in our open source repo (SDK repo ↗).

You can import our open source tools into your project using from portia.open_source_tools.registry import open_source_tool_registry and load them into an InMemoryToolRegistry object. You can also combine their use with cloud or custom tools as explained in the docs (Add custom tools ↗).

Tool details

Tool ID: llm_tool

Tool description: Jack of all trades tool to respond to a prompt by relying solely on LLM capabilities. YOU NEVER CALL OTHER TOOLS. You use your native capabilities as an LLM only. This includes using your general knowledge and your in-built reasoning. This tool can be used to summarize the outputs of other tools, make general language model queries or to answer questions. This should be used only as a last resort when no other tool satisfies a step in a task, however if there are no other tools that can be used to complete a step or for steps that don't require a tool call, this SHOULD be used. MAKE SURE the task_data includes ALL INPUT VARIABLES IN THE CONTEXT.

Args schema:

{
  "description": "Input for LLM Tool.",
  "properties": {
    "task": {
      "description": "The task to be completed by the LLM tool",
      "title": "Task",
      "type": "string"
    },
    "task_data": {
      "anyOf": [
        {
          "items": {},
          "type": "array"
        },
        {
          "type": "string"
        },
        {
          "type": "null"
        }
      ],
      "default": null,
      "description": "Task data that should be used to complete the task. Can be a string, a list of strings, or a list of objects that will be converted to strings. Important: This should include all relevant data in their entirety, from the first to the last character (i.e. NOT a summary).",
      "title": "Task Data"
    }
  },
  "required": [
    "task"
  ],
  "title": "LLMToolSchema",
  "type": "object"
}

Output schema:

('str', "The LLM's response to the user query.")