IdeaWell

Project ideas from Hacker News discussions.

Talorys – A self-hosted personal AI agent on Cloudflare's free tier

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

Theme 1 – What counts as “self‑hosted?”
Many commenters argue that running on Cloudflare’s services means you’re still relying on a third party, so the project isn’t truly self‑hosted.
- “This is a neat idea, but I wouldn’t say you aren’t relying on someone else's hosted service. Cloudflare isn’t a public utility yet.” – onionisafruit
- “Self hosted means you host it yourself… Running it in a cloud platform somewhere is stretching it but I'd say that still is. SaaS definitely is not self hosting.” – ajnin

Theme 2 – Concerns about cost, usage limits and unpredictable billing
Users worry about the free‑tier “neurons” quota, possible overages, and stories of large Durable Objects bills.
- “It seems like the free 10000 'neurons' would cover light daily usage… I spent 500 in a few minutes of chatting until I realized the tasks I had in mind aren't possible.” – tln
- “I've read yesterday how someone's DO racked $34k bill, someone's else's $17k.” – subscribed
- “Careful with Cloudflare's 'free' AI tier… I was never able to get an answer on why I was billed for their confusing neuron usage.” – hung

Theme 3 – Feedback on usefulness, architecture and desired features
Praises for the easy deployment and tech stack are paired with requests for enhancements such as web search, semantic memory, or the ability to swap in a local model.
- “This is a neat implementation … the sticking point is that there is no good implementation equivalent to the Alexa devices … that allow me to interact via speech.” – richgroot
- “It needs semantic recall though … I added a memory entry for 'I am a vegetarian' … It gave me options for chicken and salmon.” – andrewdb
- “It’s open source … it’s a relatively minor tweak to replace the AI calls to point to a local model server … less than 20 minutes of work by hand.” – vlovich123

🚀 Project Ideas

Generating project ideas…

HomeAgent

Summary

  • A self‑hosted personal AI agent that runs entirely on local hardware (e.g., Raspberry Pi, home server) using Ollama/llama.cpp for LLM inference, providing chat, task/note management, reminders, and optional remote access via Tailscale or Cloudflare Tunnel.
  • Core value proposition: true data ownership, zero reliance on third‑party AI services, and predictable performance without hidden usage costs.

Details

Key Value
Target Audience Privacy‑conscious developers, home‑lab enthusiasts, and users who want an AI assistant they fully control
Core Feature Local LLM‑powered chat with persistent memory, task/todo list, notes, and scheduled reminders
Tech Stack Ollama (or llama.cpp), Node.js/Express or Python FastAPI, IndexedDB or SQLite for persistence, optional Tailscale/Cloudflare Tunnel for remote access, Docker for easy deployment
Difficulty Medium
Monetization Hobby

Notes

  • HN commenters emphasized wanting to run on their own hardware and avoid third‑party billing surprises (e.g., “I want to run it locally in my home…”, “Self hosted means you can run it locally in your home”).
  • Provides a concrete path to the “self‑hosted” ideal discussed, while still offering simple one‑command deployment via Docker Compose.

NeuronWatch

Summary

  • A lightweight monitoring and alerting tool for Cloudflare Workers AI usage that tracks “neurons” consumption, projects future spend, and sends alerts when approaching free‑tier limits or custom thresholds.
  • Core value proposition: eliminates billing shock and helps users stay within the free tier by giving clear visibility into AI usage.

Details

Key Value
Target Audience Developers and hobbyists using Cloudflare Workers AI who worry about unpredictable billing and free‑tier exhaustion
Core Feature Real‑time usage dashboard, forecast modeling, and configurable email/push alerts
Tech Stack Cloudflare Workers (for API proxy), React/Vue frontend, Chart.js for visuals, Cloudflare KV or Durable Objects for state, optional GitHub Actions for deployment
Difficulty Low
Monetization Hobby

Notes

  • Users complained about confusing neuron usage and unexpected charges (“I got my bill reversed after having an AI agent pester them long enough”, “Their billing system truly sucks”).
  • A transparent usage tool directly addresses these frustrations and would be welcomed by the community seeking cost predictability.

SpeakAgent

Summary

  • A voice‑first personal AI assistant designed to run on low‑cost hardware (Raspberry Pi, SBC) using open‑source STT (Whisper), TTS (Piper or Coqui), and a local LLM (via Ollama/llama.cpp) for fully offline operation, with optional integration to smart‑home devices and user‑provided search APIs.
  • Core value proposition: hands‑free, private AI interaction that works without internet dependency and avoids reliance on any third‑party voice service.

Details

Key Value
Target Audience Makers, home‑automation hobbyists, and users wanting an Alexa‑like assistant they fully control
Core Feature Wake‑word detection, speech‑to‑text, LLM‑driven response, text‑to‑speech, and extensible skill system for tasks, notes, reminders, and device control
Tech Stack Raspberry Pi OS, Whisper.cpp (STT), Piper/TTS (TTS), Ollama or llama.cpp (LLM), Python or Rust for glue logic, MQTT or Home Assistant integration for smart‑home, optional user‑provided search API keys
Difficulty Medium
Monetization Hobby

Notes

  • Commenters asked for speech capabilities (“…there is no good implementation equivalent to the Alexa devices scattered around my home…”) and wanted to run on local hardware (“Self hosted means you can run it locally in your home”).
  • Provides a complete offline voice assistant, satisfying the desire for a truly self‑hosted, privacy‑preserving AI agent that can be used around the house.

Read Later