Xiaomi's open-source coding harness for LLM workflows.
Show EvidenceHide Evidence
Commenters praised the open release and argued LLM tools should be commodities, not closed platforms.
Issue / 2026-06-11
A daily board of tools, apps, and references that Hacker News readers pulled into view on 2026-06-11. Each row keeps the original HN thread close to the claim.
What surfaced that day
Xiaomi's open-source coding harness for LLM workflows.
Commenters praised the open release and argued LLM tools should be commodities, not closed platforms.
AI2's open language-model stack for reproducible training runs.
The thread said OLMo releases full datasets and is closer to a fully open pipeline than Nemotron.
NVIDIA's open model family with weights, recipes, and data.
Commenters said Nemotron exposes much of the training stack and has enough openness to serve as a good starting point.
Open reasoning-dataset project for training small reasoning models.
The thread pointed to its dataset, model releases, and curation paper as useful open infrastructure.
A fully sovereign open-source reasoning system from MBZUAI.
The project page and comments describe it as open from pre-training data through post-training and evaluation.
A network-layer firewall for AI agents.
Deno described it as a WireGuard/Tailscale firewall that parses HTTP, Postgres, and SSH and gates destructive actions.
Zed's fine-grained history layer for changes between commits.
Commenters quoted the idea that DeltaDB records every operation in between commits and argued over whether that history is useful or invasive.
A free browser-based ear-training site for musicians.
One user called it simple and great; another clarified the lock icon freezes the difficulty level.
A COBOL first-person shooter built with raycasting and PPM output.
The thread described the game pipeline and several commenters debated the AI-assisted authorship while still admiring the implementation.