how do i change this endpoint to use a grok api key instead? async def entrypoint(ctx: agents.JobContext): session = AgentSession( llm=openai.realtime.RealtimeModel.with_azure( voice="alloy", azure_deployment=os.getenv("AZURE_OPENAI_DEPLOYMENT"), azure_endpoint=os.getenv("AZURE_OPENAI_ENDPOINT"), api_key=os.getenv("AZURE_OPENAI_KEY"), api_version=os.getenv("AZURE_OPENAI_API_VERSION"),

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

is this correct: async def entrypoint(ctx: agents.JobContext): session = AgentSession( llm=openai.LLM.with_x_ai( model="grok-4-1-fast-non-reasoning", ), ) When I call the phone number I get no answer

You're close — but we need to check two things: the model/plugin choice and your telephony wiring. I’ll confirm the code and then give focused troubleshooting steps for the "no answer" phone call. Acknowledging your snippet • Your snippet using

openai.LLM.with_x_ai(...)

can be correct only if: ◦ You installed the OpenAI plugin extra (

livekit-agents[openai]

) for your livekit-agents version. ◦ You set

XAI_API_KEY

with your Grok key. ◦ You use a supported Grok model name (e.g.,

grok-2-public

or

grok-beta

), not an unsupported model string. Correct minimal example (OpenAI plugin path)

from livekit.plugins import openai
from livekit.agents import AgentSession

async def entrypoint(ctx: agents.JobContext):
    session = AgentSession(
        llm=openai.LLM.with_x_ai(
            model="grok-2-public",
            api_key=os.getenv("XAI_API_KEY"),
        ),
    )

If you intend real-time voice (low-latency speech), use the xai realtime plugin instead:

from livekit.plugins import xai

async def entrypoint(ctx: agents.JobContext):
    session = AgentSession(
        llm=xai.realtime.RealtimeModel(
            model="grok-beta",
            voice="ara",
        ),
    )

That requires

livekit-agents[xai]

and

XAI_API_KEY

set. Troubleshooting why the phone call gets no answer Please confirm or check each item below and tell me the results so I can narrow this quickly: 1. Agent process status • Is your agent running and started in dev mode (e.g.,

python agent.py dev

)? Check the agent logs for connection/registration messages. 1. Dispatch rule & agent name • Does your dispatch rule in Telephony point to the exact agent name your code registers? (Agent name must match the dispatch rule.) 1. Telephony configuration • Is the phone number associated with the dispatch rule that routes to your agent? Verify in LiveKit Cloud → Telephony → Configuration. 1. Plugin & API key • Which package did you install? Share output of

uv show livekit-agents

or

pip show livekit-agents

. • Confirm

XAI_API_KEY

is set and accessible to the running process. 1. Model choice • Which model did you specify? Use

grok-2-public

or

grok-beta

for supported flows. If using realtime voice, use

xai.realtime.RealtimeModel

. 1. Logs/errors when calling • Do you see any errors in agent logs when the call arrives (webhook/telephony logs)? Paste any error lines you see. Next step • Tell me: which plugin path you prefer (OpenAI plugin with

with_x_ai()

or xai realtime), the livekit-agents version you have, and paste any relevant log lines. I’ll provide the exact code snippet and the next fix based on that. Sources: xAI Grok Voice Agent API plugin | LiveKit docs | xAI LLM plugin guide | LiveKit docs | SIP troubleshooting guide | LiveKit docs | Simple Call Answering Agent | LiveKit docs

i have specified XAI_API_KEY in .env

Thanks — glad you confirmed

XAI_API_KEY

is set. Next, let’s narrow down why the call isn’t answered. Please provide these quick checks and answers: 1. Which plugin path are you using in code? ◦

openai.LLM.with_x_ai(...)

or

xai.realtime.RealtimeModel(...)

? 1. livekit-agents install & version (paste output): ◦ Run

uv show livekit-agents

or

pip show livekit-agents

. 1. How are you running the agent? ◦ Command you run (e.g.,

python agent.py dev

,

uv run

, etc.). 1. Agent name vs dispatch rule: ◦ Does your agent register a name? Share the

agent_name

in your code or the decorator. ◦ Confirm the Telephony dispatch rule points to that exact name. 1. Model string you set in code (exact value). 2. Any relevant logs when a call comes in (agent logs, telephony logs, or errors). Paste the most recent lines. With those details I’ll give the exact corrected snippet and the likely fix (routing, plugin mismatch, or runtime/env issue).

this is my entire entrypoint: async def entrypoint(ctx: agents.JobContext): session = AgentSession( llm=openai.LLM.with_x_ai( model="grok-4-1-fast-non-reasoning", ), )

Thanks — I can see your full entrypoint. Two immediate issues to fix so the call can be answered: 1) Model + plugin mismatch •

grok-4-1-fast-non-reasoning

is not a documented Grok model for LiveKit. Use a supported model name. • If you’re using the OpenAI plugin path, use this minimal, correct form:

from livekit.plugins import openai
from livekit.agents import AgentSession
import os

async def entrypoint(ctx: agents.JobContext):
    session = AgentSession(
        llm=openai.LLM.with_x_ai(
            model="grok-2-public",            # use a supported model
            api_key=os.getenv("XAI_API_KEY"), # ensure this is set
        ),
    )

• If you want realtime voice (low-latency), instead use the xai realtime plugin:

from livekit.plugins import xai

async def entrypoint(ctx: agents.JobContext):
    session = AgentSession(
        llm=xai.realtime.RealtimeModel(
            model="grok-beta",
            voice="ara",
        ),
    )

2) Telephony routing & runtime checks (most common cause of “no answer”) Please confirm these quickly: • livekit-agents package/version: run

uv show livekit-agents

or

pip show livekit-agents

and paste output. • Which plugin you installed:

livekit-agents[openai]

or

livekit-agents[xai]

? • How you run the agent (exact command — e.g.,

python agent.py dev

). • Agent registration name / decorator (does your agent set

agent_name

?), and that the Telephony dispatch rule points to that exact name. • Any agent logs when a call arrives (paste the last 10–20 lines). Give me those answers and I’ll provide the exact code you should run and the single next step to fix the no-answer behavior.