VX

Overview

A fast, cross-platform, GPU-accelerated desktop genome and transcriptome browser built for AI-agent driven exploration of sequencing data. It visualises FASTA, GTF, GFF, BAM, BED, BigBed, bedGraph, BigWig, VCF and BEDPE interaction tracks, runs 49 built-in analyses against them, and exposes every action through a native MCP server and an embedded HTTP API so an LLM assistant can observe, drive and analyse alongside the user.

Linux · macOS · Windows · Binaries: CC-BY-NC-ND-4.0 / MCP server + HTTP API: MIT

Quick start

curl -LO https://github.com/Arnaroo/VX/releases/download/v0.10.0/VX-0.10.0-linux-x86_64.tar.gz
tar xzf VX-0.10.0-linux-x86_64.tar.gz
./VX-0.10.0-linux-x86_64/bin/vx

What it is for

VX is a desktop genome browser in the family of IGV, the UCSC Genome Browser and JBrowse, but built for a different decade. It targets researchers working with very large sequencing datasets on a laptop, bioinformaticians who want to run analyses against the data while looking at it, AI and agentic-LLM users who want a browser their agent can see and control, and educators who need a scriptable, screenshot-and-record-ready viewer that produces publication-grade output.

Why it is different

• GPU-accelerated OpenGL level-of-detail rendering: smooth pan and zoom from chromosome scale to base level in one continuous gesture, with no tile-popping.
• Asynchronous background-thread BAM loading: the viewport never blocks, even while a multi-gigabyte alignment cache warms.
• A 58-tool MCP server plus an embedded HTTP API on 127.0.0.1:9876, so Claude, a Python script or a plain curl command can navigate, load files, capture tracks, run analyses and read their result files.
• 49 built-in analyses (peak calling, quantification, GC and CpG content, motif search, variant and coding impact, Hi-C contact and TAD boundaries, cross-track correlation) that run in the background with file output, each checked against a Python reference implementation in a public validation package.
• A manifest beside every result file: the analysis, the version, every input with its SHA-256 and the parameters as applied, so a run can be reproduced and checked.
• Built-in video and GIF recording of the OpenGL viewport for figure animations and talks.
• Self-contained binaries, with no Python runtime, Conda environment or web stack. No account and no cloud service; anonymous usage statistics only if you opt in, off by default; the API binds only to localhost.