LLMWithTools
Overview¶
The LLMWithTools Block allows interaction with a Language Learning Model (LLM) and integrates additional tools to assist the LLM in generating more complex responses. This Block supports multiple tools, which can be invoked by the LLM when needed. It manages both the LLM's responses and the results from tool invocations.
This Block is useful when you want the LLM to dynamically call different tools based on user inputs, allowing for more versatile and context-aware conversations.
Description¶
Metadata¶
- Category: Agent
Configuration Options¶
| Name | Data Type | Description | Default Value |
|---|---|---|---|
| use_thread_history | bool |
Inputs¶
| Name | Data Type | Description |
|---|---|---|
| message | list[ContentItem] or str |
Outputs¶
| Name | Data Type | Description |
|---|---|---|
| response | ResponseSchemaT |
State Variables¶
| Name | Data Type | Description |
|---|---|---|
| message | list[ContentItem] or str |
|
| tool_calls | dict[str, dict[str, Any]] |
|
| tool_messages | list[AssistantMessage or ToolMessage] |
Example(s)¶
Example 1: Use the LLM to select a tool¶
- Create an
LLMWithToolsBlock. - Define tools like
"Tool1"and"Tool2"that the LLM can choose from. - Provide an input message:
"Which tool should I use for this task?" - The Block will invoke the appropriate tool based on the LLM’s decision, process the result, and return a response.
Example 2: Handling a response without tool invocation¶
- Set up an
LLMWithToolsBlock with tools defined. - Provide a message where no tool invocation is necessary:
"Summarize the document." - The LLM will generate the response directly and send it back without calling any tools.
Example 3: Multiple tool calls in a single interaction¶
- Set up an
LLMWithToolsBlock. - The LLM can call multiple tools during the interaction, processing each tool’s output and using the results to generate a final response.
Error Handling¶
- If tool results are not available, the Block will raise an error.
- If the LLM response type is unexpected, the Block will attempt to handle the error gracefully by sending back a default response.
FAQ¶
How does the Block handle tool invocations?
The LLM can call tools defined in the tools dictionary. The Block waits for the tool's results, appends the result to the conversation history, and allows the LLM to continue processing the message.
Can I define custom tools for the LLM to use?
Yes, you can define custom tools in the tools dictionary. Each tool has its own functionality and can be invoked by the LLM during conversation based on the input message.
What happens if the LLM doesn’t need a tool?
If the LLM determines that no tools are necessary for a particular response, it will generate the response directly and return it without invoking any tools.
Can the Block handle multiple tool calls in a single interaction?
Yes, the Block is capable of handling multiple tool calls in a single interaction. It manages tool call results and appends them to the conversation history, ensuring that the LLM can utilize multiple tools as needed.