Theseus: keep only what's load-bearing

Most web agents return summaries or lists of links. This one returns a single value: a phone number, a version, a price, a date. It also cites the page the value came from, and returns an empty answer instead of guessing when the site does not contain it. A resident server loads Qwen3-4B-Instruct once onto the AMD GPU (ROCm) at container start. Each question is a fresh app.py process that writes a valid placeholder JSON immediately and then talks to the server over a unix socket, with a hard deadline under the 30-second limit. The agent crawls the start page and same-site links in parallel, ranked by how well each link matches the question. It cleans pages to text (table rows kept), ranks passages with BM25, and has the model extract the value. An answer is only accepted if it literally appears in the cited passage, which removes hallucinations. A rule-based extractor for phones, emails, versions, prices and dates is the fallback when time runs short. No API keys: all inference is local on the AMD GPU.

Necesites

An agent for teams that review npm upgrades. It tests each upgrade in a disposable Vultr sandbox, blocks hidden network calls, and returns an evidence bundle with a SHA-256 manifest.

Upgrade Proof

Every score cites its source. If the resume doesn't say it, the score is zero.

Resync

Batik to Global AI Control Tower: Design AI + OEE 95% ERP SAK EMKM + Marketing + SRITEX Distribution + GPS Live on Vultr + AMD MI300X

SuperInfra Trinity V32-2 SOLO Batik to Global