Project ideas from Hacker News discussions.

Llama.cpp on a jailbroken PS5: run any GGUF model natively on the console

📝 Discussion Summary (Click to expand)

1. Hardware feasibility & performance
- ThrowawayTestr: “11.4G of shared memory. Could do something interesting if you don’t have a GPU.”
- robotswantdata: “Significantly slower and less capable than opus 5.5 … you can run comparable or better models than gpt4. Try to stay under 30b.”
- p1necone: “They’re not even in the same ballpark – the PS5 is more than an order of magnitude better in GPU performance.”

2. Legal/takedown & corporate‑control worries
- Rabbit504030201: “The project's primary site should not be GitHub, and I am serious. Playstation 5 is not some historical curiosity and Sony might strike the page down any hour now.”
- minimaxir: “There's no feasible way to avoid takedowns/DMCA in 2026 unless you do TOR shenanigans. Even with a custom domain, Sony could just DMCA the web host for the project.”
- LoganDark: “I struggle to think of anywhere on the entire open internet you could put something like this where it could not be taken down.”

3. Vision for applications (LLM‑powered NPCs, companions, monitoring)
- pwython: “With this in mind, having LLM‑powered NPCs would be a beautiful thing. There's been limited testing on PC games (PUBG Ally), but I can imagine it will be the standard.”
- tiahura: “Fortnite Darth Vader already did.”
- greggsy: “After all of the agonising about AI companions, we realised that we just wanted someone to play games with.”
- jjcm: “Even though it only runs qwen 3.8 27b at a 2 bit quant, it's still highly viable… great for things like monitoring server error rates, reviewing security footage, watching plant growth / controlling watering…”


🚀 Project Ideas

PS5-LLM Orchestrator

Summary

  • Enables easy deployment and management of quantized LLMs on a PS5 running homebrew, providing a CLI/UI for model loading, token generation monitoring, and distributed inference across multiple PS5 nodes.
  • Turns idle PS5s into cheap AI inference nodes for background tasks such as log analysis, security‑footage review, or continuous code‑base questioning.

Details

Key Value
Target Audience Developers, hobbyists, and home‑lab enthusiasts who own a PS5 and want to run LLMs locally.
Core Feature One‑click install of a llama.cpp‑based server on PS5, auto‑quantization selection (2‑bit/4‑bit), REST API, and swarm mode to spread load across multiple PS5s.
Tech Stack llama.cpp (cross‑compiled for PS5 APU), Node.js/Express for API, libp2p for peer discovery, optional Docker for orchestration.
Difficulty Medium
Monetization Hobby

Notes

  • Commenters noted the viability of low‑quant models for monitoring: “jjcm: Even though it only runs qwen 3.8 27b at a 2 bit quant, it's still highly viable… for things like monitoring server error rates, reviewing security footage…” .
  • Provides a concrete way to exploit the PS5’s 11.4 GB shared memory and GPU for AI workloads, sparking discussion on power‑cost vs token‑throughput trade‑offs.

MagnetRepo: Decentralized GitHub Repo Archiver

Summary

  • Takes a GitHub repo URL, creates a zip, generates a magnet torrent link (and optionally an IPFS CID), and seeds it via libtorrent/IPFS.
  • Offers a simple web UI similar to Software Heritage Save but resistant to DMCA takedowns, letting users share code (including jailbreak guides) without fear of removal.

Details

Key Value
Target Audience Open‑source developers, archivists, researchers needing censorship‑resistant code distribution.
Core Feature One‑click archiving: paste repo URL → get magnet link + optional IPFS CID; background seeding continues as long as the client runs.
Tech Stack libtorrent (or WebTorrent), js-ipfs, Express backend, React frontend, SQLite for tracking seeds.
Difficulty Medium
Monetization Hobby

Notes

  • Directly addresses the request: “toomuchtodo: Any recommendations for code that’ll one shot pull a GitHub repo zip from an repo url and announce and seed a magnet torrent?” and the concern raised by Rabbit504030201 about avoiding GitHub for takedown‑prone content.
  • Enables distribution of PS5 jailbreak instructions, model weights, or any other sensitive repos, fostering debate on decentralized code hosting and DMCA resilience.

PS5 AI Companion Hub

Summary

  • Service that lets users run LLM‑powered NPCs or game companions on their PS5, with optional orchestration across multiple consoles to improve token rate.
  • Provides matchmaking, session persistence, and low‑latency voice/text integration, turning the PS5 into an interactive AI buddy for solo or co‑op play.

Details

Key Value
Target Audience Gamers, AI enthusiasts, and home‑brew PS5 owners wanting AI‑enhanced gameplay.
Core Feature Plug‑and‑play homebrew app that loads a quantized LLM, exposes a WebSocket API for game mods to send prompts/receive responses, plus a central matchmaking server to pool multiple PS5s for higher throughput.
Tech Stack llama.cpp compiled for PS5 APU, Node.js/ws WebSocket server, Unity/Unreal plugin example (C#), Redis‑based matchmaking, optional Docker orchestrator.
Difficulty High
Monetization Revenue-ready: Subscription $5/mo for premium matchmaking & model updates.

Notes

  • Reflects the desire for AI companions: “pwython: With this in mind, having LLM-powered NPCs would be a beautiful thing… limited testing on PC games (PUBG Ally)…” and “greggsy: After all of the agonising about AI companions, we realised that we just wanted someone to play games with.”
  • Could ignite discussion on token‑rate limitations, cost‑per‑hour of AI gameplay, and the emergence of AI‑driven game genres on consoles.

Read Later