Frequently Asked Questions

General

<AccordionGroup>

What is Hubify Labs?

Hubify Labs is a Scientific Discovery Platform — an IDE-like environment for running experiments, managing multi-agent AI teams, writing papers, and publishing research. Think Cursor for research instead of code.

Who is Hubify Labs for?

Researchers, scientists, and research engineers who want to:

• Run GPU-powered experiments with automated QC
• Use AI agents to accelerate analysis and paper writing
• Publish research with a public lab site
• Manage reproducibility infrastructure

It works for any research domain: cosmology, machine learning, biology, physics, and more.

Is this a hosted service or self-hosted?

Hubify Labs is a hosted service. The web UI and API run on our infrastructure. GPU compute is provisioned through your own RunPod (or Modal) account, so you control your compute costs directly.

What surfaces are available?

Three equivalent surfaces:

• Web UI at hubify.com — full-featured research IDE
• Desktop App — native macOS application (Windows coming soon)
• CLI — hubify command for terminal-first researchers

All three share the same features, data, and agents.

</AccordionGroup>

Agents

<AccordionGroup>

Which AI models are supported?

• Claude (Opus, Sonnet, Haiku) — primary models
• GPT-4o, GPT-4o-mini — for cross-model review
• Gemini 1.5 Pro — for cross-model review
• Grok-2 — for cross-model review

The orchestrator uses Claude Opus by default. Workers can use any model.

Why is cross-model review mandatory?

No single model should review its own output. Cross-model review prevents echo chambers by having GPT review Claude's work, Gemini review GPT's work, and so on. This catches model-specific biases and improves research quality.

Can I bring my own API keys?

Yes. You provide your own API keys for OpenAI, Google, and xAI. Hubify never stores these on our servers — they are configured locally in your CLI or browser.

</AccordionGroup>

Compute

<AccordionGroup>

How much does GPU compute cost?

GPU costs are billed directly by your compute provider (RunPod). Hubify does not mark up compute costs. Typical rates:

• H100: ~$2.49/hr
• H200: ~$3.89/hr
• A100: ~$1.64/hr

Set budget limits with hubify pod budget --monthly 500.

Can I use my existing RunPod account?

Yes. Connect your existing RunPod account with an API key. Hubify manages pod lifecycle but billing goes through your RunPod account.

Do I need a GPU to use Hubify?

No. Many features work without GPU compute: agent chat, paper writing, knowledge base, lab site, task management. GPU is only needed for running experiments on pods.

</AccordionGroup>

Data and Privacy

<AccordionGroup>

Where is my data stored?

• Lab metadata (experiment configs, task queues, agent settings) — Hubify servers
• Experiment outputs (chain files, figures, results) — your compute provider's storage, synced to Hubify
• Papers — stored in your lab, compiled on pods
• Chat messages — stored in your lab for continuity

Can I export all my data?

Yes. Export everything with:

hubify lab export my-lab --output ./backup.tar.gz

This includes experiments, papers, knowledge base, figures, and configuration.

Are public lab sites indexed by search engines?

Yes. Public lab sites include SEO metadata, structured data for Google Scholar, and sitemaps. Private labs are not indexed.

</AccordionGroup>

Papers

<AccordionGroup>

What LaTeX format is used?

All papers use revtex4-2 (Physical Review D style) by default. This is the standard format for physics journals. Other templates can be configured.

Can I submit directly to arXiv?

Hubify packages your paper into an arXiv-ready tarball with source, figures, and bibliography. You upload this package to arXiv manually. Direct API submission is planned for a future release.

</AccordionGroup>