Regen 1 Alpha routes every message, verifies facts against the live web, and shows you exactly how it got there — powered by GPT‑OSS‑20B on Groq.
Every message is checked against five routers in order — the first match wins. Nothing gets a generic answer without a reason.
Web search confirms the claim, or the model completes cleanly. Credits go up.
Below 700, "careful mode" kicks in — the model leans harder on evidence and flags uncertainty. Going offline or turning search off never costs credits.
Everything Regen 1 Alpha ships with, out of the box.
Runs on Groq's LPU hardware for fast, streaming-ready responses.
Optional DuckDuckGo lookups ground factual and generation queries in current sources.
Temporal, identity, generation, factual, and fallback routers — each with its own logic.
Reasoning is requested in parsed form and shown in a collapsible trace, not thrown away.
A reliability score saved to disk — survives restarts and reloads, not reset to 1000 every session.
Flask + CORS underneath. Point the frontend at your own server and run it anywhere.
What's actually running under Regen 1 Alpha.
// call your own Regen 1 Alpha backend const res = await fetch("https://your-backend.onrender.com/chat", { method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify({ message: "Summarize today's tickets", history: [], web_search_enabled: true }) }); const data = await res.json(); console.log(data.reply, data.thought, data.credits);
It runs on Groq's LPU inference hardware, which streams tokens far faster than typical GPU-hosted endpoints at this size.
No. Disabling search, or the search backend being unreachable, never deducts credits — only a genuine verification failure does.
Yes — it's a plain Flask app with CORS enabled. Point the frontend's API URL at wherever you deploy it.
"Careful mode" turns on: the system prompt asks the model to lean harder on web context and flag anything it isn't sure of.
Chat with Regen 1 Alpha directly — live web verification and visible reasoning included.