Best stack for a personal AI voice agent that calls me and walks me through packing?
这条记录涉及生成能力或端侧推理进展，适合跟踪模型效率、部署门槛和应用机会。
一位工程师希望为自己构建一个个人AI语音助手，以解决ADHD导致的旅行打包问题。该助手需能根据预设清单，通过电话与用户进行自然语音互动，包括暂停/恢复对话、追踪清单状态，并标记物品是否已打包。…
I’m an engineer trying to build a small personal tool for myself, not a business/customer-service bot.
Use case: I have ADHD and travel a lot. I want an AI voice agent that can call me before a trip and walk me through a packing list that I previously entered by text or voice.
The important part is that it needs to be interactive. I don’t just want it to read a checklist out loud. I want something like:
AI: “Do you have your passport?” Me: “Wait, let me go get it.” AI: “Okay, I’ll pause.” Me: “Got it, continue.” AI: “Next: chargers.”
Requirements:
- Outbound call to my phone based on criteria/schedule/trip timing
- Natural voice conversation
- Ability to pause/resume
- Ability to track checklist state
- Ability to mark items as packed, skipped, or still needed
- Ideally cloud-hosted, but I could run something locally on an old laptop if that is genuinely better
- Could use Twilio, Vapi, Retell, OpenAI Realtime API, Claude/Anthropic, local LLMs, or another stack
What I’m trying to figure out:
- What is the simplest current architecture for this?
- Would you use a voice-agent platform like Vapi/Retell, or build directly with Twilio + OpenAI Realtime?
- Is Claude/Anthropic useful here, or is OpenAI better because of realtime audio?
- How should I store and update checklist state during the call?
- Is local hosting worth it for this, or is cloud clearly the better choice?
- Are there any gotchas with outbound calling, latency, interruptions, or conversational state?
I’m optimizing for reliability and simplicity, not making a startup product. I just want something that actually helps me pack without needing to stare at my phone.