February 25, 2026
Factory or fluke? You decide
Show HN: Sgai – Goal-driven multi-agent software dev (GOAL.md → working code)
AI dev factory on your laptop — cheers for setup, side‑eye for license and the “Sky” name
TLDR: Sgai is a local “AI dev factory” that turns a goal file into a plan of bots that code and test your app. Commenters praised the slick one‑command setup but clashed over a “no hosted service” license clause, poked fun at the “Sky” name, and compared it to a saner Gas Town dream.
Sgai wants to be your personal “AI software factory”: you write a simple goal in a file, launch a dashboard, and a team of bots (developer, reviewer, designer) plans the work, writes code, and runs tests. The demo claims a drag‑and‑drop image compressor in 45 minutes, and the author says summer ’25 is when models finally felt “ready for autonomy.” But the community didn’t just watch the code fly — they brought the spice.
The loudest debate? A commenter flagged a non‑standard license clause that bans offering Sgai as a hosted service, turning a “modified MIT” into a “no, you can’t sell this as a cloud product” moment. Open‑source purists clutched pearls, while pragmatists shrugged: protect the project, or scare off contributors? Meanwhile, practical devs cheered the one‑command install via opencode, calling it “genius,” and asked the question everyone cares about: is this your daily driver yet?
Then came the memes. Pronunciation wars erupted over “Sgai” being said like “Sky” — one quip called it “a stretch.” And the concept itself sparked déjà vu: a commenter dubbed it a “more sane” version of Steve Yegge’s Gas Town fantasy — the automated code factory. Fans love the visual workflow and local‑only control; skeptics roll eyes at the name and the license. Either way, the factory is open — and the drama is hiring.
Key Points
- •Sgai is a local, goal-driven multi-agent system that plans, executes, and validates software tasks from a GOAL.md specification.
- •Workflows are visualized as diagrams, with agents in specialized roles (e.g., developer, reviewer, designer) and completion gated by tests or linting.
- •Users approve plans and can monitor progress, answer clarifying questions, interrupt runs, review diffs, and fork sessions.
- •Sgai runs inside the user’s repository, integrates with version control (jj recommended, Git supported), and does not auto-push to remotes.
- •Setup includes an automated opencode path and a manual install requiring Go, Node.js, bun, and opencode; the web app runs locally at http://localhost:8080.