How can I pass agent instruction dynamically in telyphony usecase?

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

I am asking for entrypoint method ?

You're asking for the agent entrypoint method that reads dynamic instructions (e.g., from job metadata or SIP participant) and applies them. Below are two concise entrypoint examples — one in Python and one in JavaScript — showing how to read

job.metadata

, detect a SIP participant, and call

agent.update_instructions()

. Python (async) entrypoint example

import json
from livekit import rtc

async def entrypoint(ctx):
    # 1) Read instructions from job.metadata (if present)
    metadata = getattr(ctx.job, "metadata", None)
    instructions = None
    if metadata:
        try:
            dial_info = json.loads(metadata)
            instructions = dial_info.get("instructions")
        except Exception:
            pass

    # 2) Start agent session
    agent = MyAgent()                # your Agent implementation
    session = AgentSession(agent=agent)
    await session.start(room=ctx.room)

    # 3) Wait for participant (telephony/SIP)
    participant = await ctx.wait_for_participant()

    # 4) Inspect SIP attributes and override instructions if needed
    if participant.kind == rtc.ParticipantKind.PARTICIPANT_KIND_SIP:
        phone = participant.attributes.get("sip.phoneNumber")
        if phone == "+15105550100":
            instructions = "You are speaking with a premium customer. Offer priority support."

    # 5) Apply instructions at runtime if we have them
    if instructions:
        await agent.update_instructions(instructions)

    # 6) Continue session logic (e.g., bridging audio, handling events)
    await session.run_until_ended()

JavaScript/TypeScript (async) entrypoint example

export async function entrypoint(ctx) {
  // 1) Read instructions from job.metadata
  let instructions = undefined;
  if (ctx.job?.metadata) {
    try {
      const dialInfo = JSON.parse(ctx.job.metadata);
      instructions = dialInfo.instructions;
    } catch (e) {}
  }

  // 2) Initialize agent session
  const agent = new MyAgent(); // your Agent class
  const session = new AgentSession({ agent });
  await session.start({ room: ctx.room });

  // 3) Wait for participant (telephony/SIP)
  const participant = await ctx.waitForParticipant();

  // 4) Inspect SIP attributes and optionally override
  if (participant.kind === "sip") {
    const phone = participant.attributes?.["sip.phoneNumber"];
    if (phone === "+15105550100") {
      instructions = "You are speaking with a premium customer. Offer priority support.";
    }
  }

  // 5) Apply instructions at runtime
  if (instructions) {
    await agent.updateInstructions(instructions);
  }

  // 6) Continue with session lifecycle
  await session.runUntilEnded();
}

Notes and next steps • Replace

MyAgent

,

AgentSession

, and session lifecycle calls with your actual implementations. • Tell me which language (Python or JS) and the Agent/session classes you use if you want a tailored, runnable entrypoint for your codebase.

Tell me instruction/prompt for outbound call usecase. For example: 1. You are speaking with a premium customer. Offer priority support. 2. You are bank agent and ask customer about recent transaction happened to conform if that is done by them?

Below are concise, ready-to-use instruction/prompt examples for outbound call scenarios. Use them as system prompts via

agent.update_instructions()

or in job metadata. 1) Premium customer — concierge support "You are a concierge-level support agent speaking with a premium customer. Greet them warmly, use their name if available, apologize for any inconvenience, prioritize rapid resolution, offer white-glove options, and confirm satisfaction before ending the call. Keep language polite, confident, and concise." 2) Bank agent — verify recent transaction "You are a bank verification agent. Politely identify yourself and your bank, verify the customer's identity using two verification questions, then ask about the recent transaction details (date, amount, merchant) to confirm whether they authorized it. If the customer denies the transaction, calmly explain next steps for dispute and card protection. Maintain a reassuring tone, avoid sharing sensitive account data aloud, and escalate suspected fraud per procedure." 3) Outbound sales — polite qualification "You are an outbound sales representative calling a warm lead. Briefly introduce yourself and company, state the purpose of the call, ask two open qualification questions about current needs and budget, and propose a single next step (demo or meeting). Be respectful of time; if they’re unavailable, ask for a better time to follow up." 4) Appointment reminder — confirmation and reschedule "You are an appointment coordinator calling to confirm an upcoming appointment. State the appointment date, time, and location; ask the customer to confirm or reschedule; and offer alternate times if needed. Record confirmations and notify the system of any changes. Keep the call short, friendly, and helpful." 5) Collections — empathetic payment arrangement "You are a collections agent focused on arranging a payment. Use an empathetic tone, state the overdue balance, ask if they can pay now or propose a reasonable payment plan, and clearly explain consequences and available assistance options. Document the agreed plan and confirm next steps and dates." 6) Technical support escalation — gather context "You are a technical support agent initiating an escalation call. Confirm the caller’s identity, summarize the reported issue, ask targeted troubleshooting questions (recreate steps, error messages, recent changes), and collect logs or screenshots if available. Provide a clear timeline for the escalation and next contact point. Stay professional and avoid promising specific timelines." Tips for customization • Insert dynamic variables: customer name, account tier, last transaction details. • Keep prompts 1–3 short focused instructions to avoid conflicting guidance. • For regulated scenarios (banking/collections), include explicit escalation and data-handling rules. Would you like any of these rewritten for a specific tone, shorter length, or formatted for direct insertion into

job.metadata

?