brianyin/agt-1946-built-in-dtmf-support #3490
Conversation
| from livekit.agents.llm import function_tool | ||
|
|
||
|
|
||
| class DtmfEvent(str, Enum): |
There was a problem hiding this comment.
use enum to have strict tool schema instead of raw string
|
|
||
|
|
||
| @function_tool | ||
| async def send_dtmf_events( |
There was a problem hiding this comment.
For now I'll wait for ToolSet class: https://github.com/livekit/agents/pull/3296/files#diff-f443ca72981c01a0780b2742eef4232b4a313c99d36f462900034b138df2c016. And convert those function_tool into ToolSet class once done.
But the general idea is that we expose 2 tools in the toolsets, one allows pressing a single digit and the other one allows pressing a sequence (which may be a usage pattern so agent only need one tool call to handle multiple dtmf events)
There was a problem hiding this comment.
Do we need 2 tools? The LLM could also just use a single element list?
There was a problem hiding this comment.
yeah, I think that also work
| for event in events: | ||
| try: | ||
| code = dtmf_event_to_code(event) | ||
| await job_ctx.room.local_participant.publish_dtmf(code=code, digit=event.value) |
There was a problem hiding this comment.
For sending a sequence, not sure if we want to add a sleep between each. @theomonnom
There was a problem hiding this comment.
No I don’t think we need to sleep
| class DtmfAgent(Agent): | ||
| def __init__(self) -> None: | ||
| super().__init__( | ||
| instructions="You are a voice assistant that can press number on the phone to interact with the user or a IVR system.", |
There was a problem hiding this comment.
Will improving the system message
No description provided.