界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
效率与协作 · SKILL.md

retellai-sdk-patterns

retellai-sdk-patterns

【效率与协作】Production-ready Retell AI SDK patterns for voice agent applications. Use when building production voice agents, implementing retry logic, or establishing patterns. Trigger with phrases like "retell patterns", "voice agent patterns", "retell best practices".

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: retellai-sdk-patterns description: 'Production-ready Retell AI SDK patterns for voice agent applications.

Use when building production voice agents, implementing retry logic, or establishing patterns.

Trigger with phrases like "retell patterns", "voice agent patterns", "retell best practices".

' allowed-tools: Read, Write, Edit version: 1.9.0 license: MIT author: Jeremy Longshore jeremy@intentsolutions.io tags:

  • saas
  • retellai
  • voice
  • telephony
  • patterns compatibility: Designed for Claude Code

Retell AI SDK Patterns

Overview

Production-ready patterns for Retell AI: client singletons, typed agent configurations, call management, and error handling.

Prerequisites

  • Completed retellai-install-auth
  • retell-sdk installed

Instructions

Step 1: Singleton Client

import Retell from 'retell-sdk';

let _retell: Retell | null = null;

export function getRetellClient(): Retell {
  if (!_retell) {
    _retell = new Retell({ apiKey: process.env.RETELL_API_KEY! });
  }
  return _retell;
}

Step 2: Typed Agent Configuration

interface AgentConfig {
  name: string;
  voiceId: string;
  prompt: string;
  functions?: FunctionConfig[];
  maxCallDurationMs?: number;
  endCallAfterSilenceMs?: number;
}

async function createAgent(config: AgentConfig) {
  const retell = getRetellClient();

  const llm = await retell.llm.create({
    model: 'gpt-4o',
    general_prompt: config.prompt,
    functions: config.functions,
  });

  const agent = await retell.agent.create({
    response_engine: { type: 'retell-llm', llm_id: llm.llm_id },
    voice_id: config.voiceId,
    agent_name: config.name,
    max_call_duration_ms: config.maxCallDurationMs || 300000,
    end_call_after_silence_ms: config.endCallAfterSilenceMs || 10000,
  });

  return { agentId: agent.agent_id, llmId: llm.llm_id };
}

Step 3: Call Manager with Retry

async function makeCallWithRetry(
  fromNumber: string, toNumber: string, agentId: string, maxRetries = 2
) {
  const retell = getRetellClient();
  for (let attempt = 0; attempt <= maxRetries; attempt++) {
    try {
      const call = await retell.call.createPhoneCall({
        from_number: fromNumber,
        to_number: toNumber,
        override_agent_id: agentId,
      });
      return call;
    } catch (err: any) {
      if (attempt === maxRetries || err.status < 500) throw err;
      await new Promise(r => setTimeout(r, 2000 * (attempt + 1)));
    }
  }
}

Step 4: Batch Call Campaign

async function runCallCampaign(
  numbers: string[], agentId: string, concurrency = 3, delayMs = 2000
) {
  const results: Array<{ number: string; callId?: string; error?: string }> = [];
  const queue = [...numbers];
  const active = new Set<Promise<void>>();

  while (queue.length > 0 || active.size > 0) {
    while (active.size < concurrency && queue.length > 0) {
      const number = queue.shift()!;
      const p = (async () => {
        try {
          const call = await makeCallWithRetry(process.env.RETELL_PHONE_NUMBER!, number, agentId);
          results.push({ number, callId: call.call_id });
        } catch (err: any) {
          results.push({ number, error: err.message });
        }
        await new Promise(r => setTimeout(r, delayMs));
      })();
      active.add(p);
      p.finally(() => active.delete(p));
    }
    if (active.size > 0) await Promise.race(active);
  }
  return results;
}

Output

  • Singleton Retell client
  • Typed agent creation with LLM configuration
  • Retry logic for call creation
  • Concurrent call campaign manager

Error Handling

PatternUse CaseBenefit
SingletonAll SDK callsOne client instance
Typed configAgent creationType safety, defaults
Retry wrapperCall failuresAutomatic recovery
Campaign managerOutbound callsRate-limited concurrency

Examples

Use an idempotency boundary around call creation

Generate a request identifier at the application edge and persist its state before invoking the Retell SDK. If the network response is lost, retrieve the existing result by that identifier or operator-visible correlation data before attempting another call. Keep retries bounded and limited to operations whose outcome can be established; this prevents a transient client failure from creating multiple customer calls.

Resources

Next Steps

Apply in retellai-core-workflow-a for agent building.

我的收藏