I am getting this error with groq - livekit.agents._exceptions.APIStatusError: Error code: 400 - {'error': {'message': "invalid JSON schema for tool end_call, tools[2].function.parameters: 'required' present but 'properties' is missing", 'type': 'invalid_request_error', 'param': 'tool end_call, tools[2].function.parameters'}} (status_code=400, request_id=req_01kkhgfnjvfxj9mrjp0der44sp, body={'message': "invalid JSON schema for tool end_call, tools[2].function.parameters: 'required' present but 'properties' is missing", 'type': 'invalid_request_error', 'param': 'tool end_call, tools[2].function.parameters'}, retryable=True)

Thanks for reaching out, Harsh. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

how should i change the function call then - This is my end call function -

@function_tool()
async def end_call(self,
                   ctx: RunContext):
    """Called when the agents has collected all information and is ready to end the call."""
    <http://log.info|log.info>(f"[{self.agent_name}][correlation_id: {self.room_name}]: Ending the call")
    # await self.session.say(text=f"{message}")
    await ctx.wait_for_playout()
    await asyncio.sleep(0.5)
    self.disconnectedBy = 'Bot'
    await self.hangup()

Thanks — I can see your

end_call

implementation is an argument-less async function, which means your tool schema should either declare no required parameters or include an empty

properties

object. The JSON schema validation error happened because

required

was present in the tool definition but

properties

was missing. Two ways to fix this depending on how you register the tool: 1) If you register the tool with no parameters (recommended here) • Remove any

required

from the tool's

parameters

schema and set

properties

to an empty object. Example using

raw_schema

(or the decorator that accepts schema):

@function_tool(raw_schema={
  "name": "end_call",
  "description": "End the current call.",
  "parameters": {
    "type": "object",
    "properties": {}
  }
})
async def end_call(self, ctx: RunContext):
    # your existing implementation
    <http://log.info|log.info>(f"[{self.agent_name}][correlation_id: {self.room_name}]: Ending the call")
    await ctx.wait_for_playout()
    await asyncio.sleep(0.5)
    self.disconnectedBy = 'Bot'
    await self.hangup()

2) If you want the tool to accept parameters (for example a

reason

) • Define

properties

and

required

to match the expected fields. Example:

@function_tool(raw_schema={
  "name": "end_call",
  "description": "End the current call.",
  "parameters": {
    "type": "object",
    "properties": {
      "reason": {
        "type": "string",
        "description": "The reason for ending the call"
      }
    },
    "required": ["reason"]
  }
})
async def end_call(self, ctx: RunContext, reason: str = None):
    <http://log.info|log.info>(f"[{self.agent_name}][correlation_id: {self.room_name}]: Ending the call: {reason}")
    await ctx.wait_for_playout()
    await asyncio.sleep(0.5)
    self.disconnectedBy = 'Bot'
    await self.hangup()

Which option fits you? • If

end_call

should have no arguments, use option 1. • If you need arguments (like

reason

), use option 2 and adjust the function signature to accept them. If you share how you currently register the tool (the decorator call or the tool list where

end_call

is defined), I’ll provide the exact corrected snippet to paste in.

So if i have to use any of the Open AI models, I have to follow this way for defining function_tools

You're on the right track — OpenAI models require function definitions to include a valid JSON schema when you register them for function calling. That means when you expose a Python function as a

function_tool

(or otherwise register it as a callable tool for the model), its

parameters

schema must be a valid JSON Schema object. Concretely: • If you include a

required

array, you must also include a

properties

object that lists the required fields. • If the function takes no arguments, omit

required

and either include an empty

properties

object or no

parameters

block at all (depending on your tool registration method). What this looks like in practice • No-argument tool (recommended for your current

end_call

):

@function_tool(raw_schema={
  "name": "end_call",
  "description": "End the current call.",
  "parameters": {
    "type": "object",
    "properties": {}
  }
})
async def end_call(self, ctx: RunContext):
    # your implementation

• Tool with arguments (only if you need them):

@function_tool(raw_schema={
  "name": "end_call",
  "description": "End the current call.",
  "parameters": {
    "type": "object",
    "properties": {
      "reason": {
        "type": "string",
        "description": "The reason for ending the call"
      }
    },
    "required": ["reason"]
  }
})
async def end_call(self, ctx: RunContext, reason: str):
    # handle reason

Quick checklist to avoid the JSON schema error you saw • Do not list fields in

required

unless they are defined under

properties

. • Ensure

parameters

is an object with

"type": "object"

when using properties/required. • If you register tools in a list (not via decorator), check that each tool’s

function.parameters

follows the same rules. If you want, paste the decorator or the tool registration snippet you currently use (the

@function_tool(...)

call or the tool list) and I’ll give the exact corrected schema to paste in.