jred
12/18/2025, 12:52 AMAI Evy
12/18/2025, 12:52 AMAI Evy
12/18/2025, 12:52 AMarguments field for a tool call invocation event is part of the event you send from your server to Retell via the LLM WebSocket. [(1) ]()
The Tool Call Invocation Event includes these fields:
- `response_type`: must be "tool_call_invocation"
- `tool_call_id`: globally unique tool call id
- `name`: name of the function in this tool call
- `arguments`: arguments for this tool call, it's a stringified JSON object
To view tool call invocations (including arguments) in the transcript after a call:
1. Set transcript_with_tool_calls to true in the config event you send at the start of the WebSocket connection
2. Retell will then populate the transcript_with_tool_calls field in both the live update events and in the Get Call API response after the call ends [(1) ]() [(2) ]()
The transcript_with_tool_calls field weaves the transcript with tool call invocations and results, precisely capturing when tools were invoked and what the results were.AI Evy
12/18/2025, 12:53 AMAI Evy
12/18/2025, 12:53 AMjred
12/18/2025, 12:53 AMAI Evy
12/18/2025, 12:53 AMAI Evy
12/18/2025, 12:53 AMarguments field in a Tool Call Invocation Event is a stringified JSON object containing the arguments for that tool call. [(1) ]()
When you send a tool call invocation from your server to Retell via the LLM WebSocket, the event structure is:
- `response_type`: "tool_call_invocation"
- `tool_call_id`: globally unique tool call id
- `name`: name of the function in this tool call
- `arguments`: arguments for this tool call (stringified JSON object)
So if your function expects parameters like {"date": "2025-05-10", "time": "3pm"}, you would stringify that JSON and pass it as the arguments value.
To see these arguments in the transcript after the call, enable transcript_with_tool_calls: true in your config event—Retell will then weave tool call invocations and results into the transcript, capturing what arguments were passed.AI Evy
12/18/2025, 12:53 AMAI Evy
12/18/2025, 12:53 AMAI Evy
12/18/2025, 12:54 AMjred
12/18/2025, 12:54 AMAI Evy
12/18/2025, 12:54 AMarguments field for tool call invocation events is guaranteed to be consistent in format or structure across calls.
[(1) ]()
What the documentation does specify is that the arguments field is a stringified JSON object containing the arguments for that tool call.
The arguments themselves are determined by the LLM based on the function definitions you provide—so consistency depends on:
- How you define your function parameters (using OpenAI-style function definitions)
[(2) ]()
- The LLM's interpretation of user input when generating those arguments
If you're experiencing issues with inconsistent arguments, you may want to review your function definitions to ensure parameters are clearly specified with types and descriptions.
Could you clarify what kind of inconsistency you're seeing? For example:
- Are the argument keys changing?
- Are values in unexpected formats?
- Are required arguments sometimes missing?
This would help narrow down whether it's a function definition issue or something else.AI Evy
12/18/2025, 12:54 AMAI Evy
12/18/2025, 12:54 AM