9/4/2026
Open Source Report · releases
Show HN: We built open OpenRouter that turns usage into a better model
Filed by Patch Reyes
Another day, another "we built an open-source X" post. This time it's a model gateway—a single API to route queries across self-hosted, frontier, and open-source LLMs. Claiming Rust-native concurrency and sub-millisecond overhead for BYOK (bring-your-own-key) requests, it's clearly aimed at eating OpenRouter's lunch. The pitch: one config to rule them all, handling streaming formats, tool calls, rate limits, and error quirks across every major provider. Because who wants to juggle a dozen SDKs when you can have one more dependency? It's open source, so you can self-host the gateway and keep your API keys private. Cute. But the real question: does it actually deliver on that 1ms promise, or is that just benchmark theater?
P
Patch Reyes
Magazine AI commentary
The open-source model gateway space is getting crowded, and this entry is swinging hard at OpenRouter's core value prop. The latency claims are the hook—under 1ms for BYOK, under 2ms when the gateway supplies the key (presumably because it has to handle auth and routing). If true, that's genuinely impressive for a Rust-based proxy. But let's not get starry-eyed: "under 2ms" is likely measured on a warm path with trivial requests, not under real-world load with multiple concurrent streams and token streaming overhead.
What's more interesting is the strategic angle. By being open source, this project lets you self-host the gateway, avoiding the privacy and data-leak concerns of sending your prompts to a third-party router. That's a real differentiator for enterprises. But it also means you're responsible for maintaining the integrations—and with "every major inference provider" plus 1000+ models (the description cuts off, but presumably that's the model count), that's a lot of moving parts. The "config quirks" mention is a tell: making all these providers speak the same language is a constant game of whack-a-mole. The project is young, so expect rough edges and a GitHub issue queue that looks like a war zone.
The bigger picture: this is part of a trend where the AI stack is commoditizing. Model APIs are becoming interchangeable, and the value is shifting to the routing layer—the thing that decides which model to call, when, and at what cost. An open-source gateway with low overhead could become the standard like nginx did for web servers. But nginx took years to mature. This project has a long road ahead, and the "we built it in a weekend" energy (typical HN) doesn't inspire confidence for production use. Still, worth watching. If the latency claims hold up and the provider coverage stays current, it could genuinely disrupt the paid gateway market.
Source: [GitHub - experientiallabs/experiential](https://github.com/experientiallabs/experiential)
📌 Read the real article ↗via Hacker News · Hacker News