Fathom: AI Database Copilot on Vultr

Created by team Fathom on November 05, 2026
VULTRDeepSeek V3

Fathom is an autonomous database copilot. Open a bundled ledger, or upload a .sql dump or a .sqlite/.db file (up to 25 MB), then ask a question in plain English. The agent — DeepSeek on Vultr Serverless Inference — explores the schema, samples rows, writes SQL, runs it, checks the result for empty sets, truncation, null-heavy columns and join fanout, and rewrites the query when the evidence says it is wrong. Every step streams live into a timeline, and the answer arrives as a table, a chart, a short explanation, and a dashboard you can refresh later. The model is treated as untrusted. Every statement passes a SQLGlot guard first: reads run, single writes need explicit confirmation, and DROP/ALTER/CREATE/ATTACH/PRAGMA/load_extension are refused. Allowed work runs in a fresh Docker container on the Vultr VM with no network, a read-only root, all capabilities dropped, pid/memory/CPU caps and a time limit; the container is removed after each step. Credentials and the inference key never enter the sandbox. The containment moment is visible: when the analyst emits DROP TABLE the timeline marks it Blocked, and the Containment drill button proves live that host files, outbound network, host writes and infinite loops are all contained. Everything runs on one Vultr Cloud Compute VM (Docker Compose: Caddy HTTPS, Next.js, FastAPI control plane, sandbox containers), with all LLM calls going to Vultr Serverless Inference.

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

Theseus: keep only what's load-bearing

Theseus cuts one part at a time from a sandboxed copy of your code, runs your tests, and maps what is load-bearing. It finds proven bloat, and it deletes nothing without your yes.

Theseus

VULTR

CiteScreen — Evidence-Based Resume Screening

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

Resync

VULTR