February 6, 2026
OSS or SOS? Pick your fighter
Show HN: Horizons – OSS agent execution engine
AI agent toolkit lands: fans hype, skeptics ask scale, license cops yell “not open”
TLDR: Horizons dropped a platform to run and evaluate AI assistants with memory, prompt tuning, and human checks. Commenters split between eager adopters and scale skeptics, while another called it “not really open source” because of a license that restricts competitors and becomes fully open after two years.
Show HN lit up as [Horizons] launched an AI agent “control tower”—a toolkit to run smart helpers with memory, prompt tuning, scorecards, and human approvals. It ships Rust under the hood plus Python/TypeScript kits, and even a one‑command local server. The pitch: make agents work in real production, not just demos.
Fans cheered. One team said they’re “moving past manual prompt optimization” and likely adopting Horizons, calling the API “surprisingly rich.” Cue the skeptics: another voice asked the million‑dollar question—how big must you be for this to matter? Translation: do you need a whole airport before you have more than one drone? That sparked a mini war between “build the runway now” and “start with a dirt road.”
Then the license sirens blared. A commenter dropped the hammer: “Not really OSS though.” Horizons uses the Sentry‑style Fair Source License (FSL): free to use and modify, but not to build a competing product. It converts to Apache 2.0 (fully open) after two years. The thread instantly turned into a meme parade: “Open source—ish,” “OSS with training wheels,” and a countdown timer GIF to 2028.
In short: hype, hesitation, and license drama. The agents are smart; the comments were smarter—and way spicier
Key Points
- •Horizons is a production platform for AI agents with event routing, long-term memory, prompt optimization, evaluation, and human-in-the-loop approvals.
- •The system comprises multiple Rust crates (horizons_core, horizons_rs, horizons_integrations, voyager, mipro_v2, rlm), all using Rust edition 2024.
- •Python and TypeScript SDKs (v0.0.6) are provided; local setup is available via Docker Compose or source build with Rust 1.85+.
- •Architecture includes pub/sub events with glob matching and DLQ, context refresh from external sources, agent review policies, connectors (Jira, LinkedIn), and queue backends (SQS, RabbitMQ).
- •The project is licensed under FSL-1.1-Apache-2.0, restricting competing-product use and converting to Apache 2.0 after two years.