Hacker News

Show HN: M-Courtyard – Fine-tune LLMs on your Mac with zero code

Hacker News - Mon, 02/16/2026 - 10:52pm

Hi HN! I built M-Courtyard because fine-tuning LLMs locally was frustrating — CLI tools, Python scripts, and scattered documentation.

This app wraps the full pipeline (document → dataset → fine-tune → test → export to Ollama) into a single desktop app that runs entirely on Apple Silicon.

Tech: Tauri 2.x + React + mlx-lm + Ollama License: AGPL 3.0

Would love feedback on what models or features would be most useful.

Comments URL: https://news.ycombinator.com/item?id=47043508

Points: 1

# Comments: 1

Categories: Hacker News

Ask HN: How do companies that use Cursor handle compliance?

Hacker News - Mon, 02/16/2026 - 10:48pm

I'm trying to decide whether to adopt Cursor for our company, but we're in a heavily regulated industry and our compliance team is flagging concerns about HIPAA/SOC2/audit trails.

The thing is, there are companies in regulated industries using it [1][2]. But Cursor has no HIPAA BAA, no FedRAMP certification, and is cloud-only with all requests routing through their AWS infrastructure. (This is probably true for Claude and other coding assistants, though I've only looked seriously at Cursor.)

So how are regulated companies actually making this work? Or do most just avoid Cursor and other AI coding tools altogether?

[1] 165 healthcare companies use Cursor according to Bloomberry: https://bloomberry.com/data/cursor/

[2] Cursor's customers include Sanofi, Johnson & Johnson, and Neuralink: https://cursor.com/customers

Comments URL: https://news.ycombinator.com/item?id=47043484

Points: 2

# Comments: 1

Categories: Hacker News

Pain in Invertebrates

Hacker News - Mon, 02/16/2026 - 10:47pm
Categories: Hacker News

The AI Body Problem

Hacker News - Mon, 02/16/2026 - 10:47pm
Categories: Hacker News

Show HN: Stop Losing LangGraph Progress to 429 Errors

Hacker News - Mon, 02/16/2026 - 10:01pm

Hey HN, I built this because I kept losing progress in LangGraph workflows when OpenRouter or OpenAI returned 429s. The problem: You're 7 steps into an agent workflow. Step 7 hits a rate limit. Everything crashes. Restart from step 1. Client-side retries don't help at scale:

100 workers all retry independently → retry storm Sequential fallbacks are slow (try OpenRouter, wait 5s, try Anthropic, wait 5s) No coordination across instances

So I built a coordination layer that:

Races multiple providers simultaneously (OpenRouter + Anthropic + OpenAI) Coordinates retries across all workers (no retry storms) Resumes workflows via webhooks (idempotent keys = checkpoints)

It runs on Fly.io's anycast network + BEAM for distributed coordination. Architecture deep dive: https://www.ezthrottle.network/blog/making-failure-boring-ag... Happy to answer questions about the approach or why BEAM made this possible when other languages would struggle.

Comments URL: https://news.ycombinator.com/item?id=47043197

Points: 1

# Comments: 0

Categories: Hacker News

Show HN: Purely Vibe Coded Asmongold Simulator

Hacker News - Mon, 02/16/2026 - 9:51pm

I hate AI (as it currently is and currently is used) and I've avoided doing a purely vibe coded project for a long time. Any time I even attempted to vibe code for a few minutes, I immediately felt disgust.

For research purposes, I gave this a proper go and created an Asmongold Simulator game.

My opinion hasn't changed on "vibe coding"

Comments URL: https://news.ycombinator.com/item?id=47043149

Points: 1

# Comments: 0

Categories: Hacker News

Pages