Theseus: keep only what's load-bearing

Created by team Theseus on November 05, 2026
VULTR

Teams keep dead code because deleting it feels risky. Nobody knows what the tests really rely on. Theseus makes that risk visible, with proof. It copies your project and never writes the original. It cuts one part at a time from the copy and runs your tests inside a bwrap sandbox: no network, no host home, no keys, read-only system, time and memory caps. If the tests break, the part goes back and is marked load-bearing. If they pass, it is not needed. It is proven bloat only when tests cover it. A blast radius map shows every part. Approve writes a new trimmed copy and re-runs the tests. Nothing is deleted without your yes. An opt-in agent on Vultr Serverless Inference plans the cut order, explains the biggest breaks in plain words and writes the brief. The tests back each call. Real runs on a Vultr VM (6 vCPU, 12 workers), v0.2.3: - expressjs/compression: 36 parts in 52.9 s. 24 load-bearing, 1 proven bloat (the Buffer import, index.js line 18), 11 flagged for a human. - expressjs/express: 402 parts. 354 load-bearing, 48 not needed, 0 proven bloat. All 402 verdicts matched a second machine. - Agent cost: about $0.003 to $0.006 per scan. Containment is shown live: hostile code tries the network, the home folder, SSH and Vultr keys, /usr, 600 processes and 2 GB. Every attempt fails.

Category tags:
GithubPresentationDemo
Explore more applications

Exact Web Answer Agent on AMD ROCm

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

AMD ROCmQwen3AMD Developer Cloud

Upgrade Proof

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

Qwen3VULTR

CiteScreen — Evidence-Based Resume Screening

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

Resync

VULTR
Streamlit
application badge

SAAS Fashion Garment Marketing - Batik to Global

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

AMD ROCmGithub CopilotHuggingFace HubLLaMALLM GatewayMongoDBStreamlitVULTR