v4 · bloomberg
HomeFeaturesLabsPricingDocsGuidesBlogSign inGet started
SCIENTIFIC DISCOVERY PLATFORM

The platform for
scientific discovery.

Thousands of datasets. Multi-agent multi-model peer review. GPU compute on demand. Make real discoveries, not summaries — in the brief window before the moats close.

Experiments
53
Anomalies
328K
Papers
4
Surveys
15
Start your discoveryView live demofree for 1 lab · no credit card · agents on call from minute one
app.hubify-labs.com / bigbounce
Hubify Labs
BigBounce
Current Lab
Captain
Projects 5
Experiments 53
Surveys 15
Papers 4
Figures 22
Data 15
Knowledge 16
Agents 11
HGhouston
Orchestrator
HG
Run SPHEREx f_NL forecast with 5-tracer config
BB
Running. sigma=0.93, 4.7 sigma detection
Ask about your research...
CaptainExperimentsPapers
Experiments
53
Anomalies
328K
Papers
4
Surveys
15
f_NL4.7σ
Bire20.7σ
PTABF 8.54
Recent Experiments
SPHEREx f_NLSPHERExDone7/7
NaMaster BirePlanckDone7/7
Combined PTANANOGravDone7/7
DESI AnomalyDESIDone7/7
ZTF DR21ZTFRun3/7
WEB
full IDE
in the browser
DESKTOP
full IDE
native macOS app
CLI · TUI
full IDE
in your terminal
all 3 surfaces stay in sync · always-on infrastructure underneath keeps your discoveries moving 24/7
bigbounce lab·53 experiments·4 papers·328K anomalies·16 contributions·day 218 active
THE OPPORTUNITY · 2025 – 2027 WINDOW
“There's a window between 2025 and 2027 where a single independent researcher can do work that beats whole institutions. The AI tooling is good enough for a 50–100x leverage multiplier, but the platforms haven't been captured yet. After 2027, the moats start closing.”
~32months until the moats close
The window is closing. Make your discovery now.
free for 1 lab · no credit card · agents on call from minute one
WHAT YOU CONNECT TO

Thousands of datasets. Every domain in science.

Hubify Labs ships with first-class connectors to the major scientific data sources. HuggingFace · NASA · arXiv · Wolfram · the K-Dense 250-database catalog · plus the long tail of domain-specific archives. Pull data. Train custom models. Publish enhanced catalogs back. Real discoveries, not summaries.

10,000+
datasets
across HuggingFace · NASA · arXiv · Wolfram · ESO · NOAA · NIH · the K-Dense catalog · and the long tail.
250+
live database connectors
PubMed · ChEMBL · UniProt · SDSS · Gaia · DESI · SEC EDGAR · FRED · BioPython · BioServices · and more.
200+
scientific data formats
FITS · HDF5 · BAM · VCF · mzML · DICOM · CIF · GeoTIFF · NWB · Zarr · Parquet · and the rest of the zoo.
14
scientific domains
Genomics · astronomy · chemistry · materials · neuroscience · imaging · geospatial · and beyond.
HuggingFace
1.5M+ models
NASA
MAST · ADS · IPAC
arXiv
2.4M+ papers
Wolfram
Data Repository
PubMed
36M+ citations
ChEMBL
2.4M molecules
UniProt
250M proteins
Gaia DR3
1.8B stars
SDSS
2.3M spectra
DESI DR1
22.5M spectra
LAMOST
11.4M spectra
eROSITA
930K sources
Mat. Project
150K crystals
NOAA
climate · ocean
SEC EDGAR
filings · finance
+ 235 more
long-tail archives
Genomics & Sequencing
Sequence & Phylogenetics
Chemistry & Molecular
Materials Science
Imaging & Pathology
Mass Spectrometry
Astronomy
Neuroscience
Single-Cell & Arrays
Geospatial & Climate
Cosmology
Particle Physics
Finance & Macro
Documents & Outputs
Built on the K-Dense scientific skills baseline · extended with HuggingFace · arXiv · Wolfram · NASA archives · and 235 more long-tail connectors.
WHO RUNS IT FOR YOU

11 agents. 4 model providers. Zero echo chamber.

Every lab ships with a pre-wired team of 11 agents: 1 orchestrator, 4 domain leads, and 6 workers — plus 4 cross-provider reviewers from different model families. You stay in the driver's seat. They handle the boring parts — and they argue with each other before they show you anything.

PI
1 · the human
You. Sets direction, approves work, owns judgement calls.
You
independent researcher
in the driver's seat · always
Orchestrator
1 · lab brain
Per-lab orchestrator. Routes work to leads · synthesizes results · runs 24/7 on Fly.
Lab Orchestrator
your lab · your leads
Anthropic · Claude Opus 4.6
Leads
4 · domain owners
Specialized leads. Each owns a research domain end-to-end.
Paper Lead
manuscripts · figures
Claude Sonnet 4.6
Experiment Lead
dispatch · pipelines
Claude Sonnet 4.6
Anomaly Lead
discoveries · novelty
Claude Sonnet 4.6
GPU Manager
routing · costs · pods
Claude Sonnet 4.6
Workers
6 · executors
Parallel task runners. Cheap, fast, focused. Always-on.
DataLoader
ingest · cleanup
Claude Haiku 4.5
Coder
scripts · pipelines
Claude Haiku 4.5
Analyst
stats · viz
Claude Haiku 4.5
Wiki-Worker
knowledge graph
Claude Haiku 4.5
+ 2 more
specialized executors
mixed Haiku · Sonnet
Cross-Provider Peer Review
4 reviewers · 4 model families · zero echo chamber

Every meaningful result is adversarially reviewed by 4 agents from different providers before it reaches you. Each model family has different blind spots, different priors, and different failure modes — so they catch what the others miss. No single model gets to be the only voice in the room.

OpenAI
GPT-5
methodology reviewer
Catches ill-posed experiments and statistical sins. Asks “would this hold if we changed the prior?”
Google
Gemini 2.5 Pro
long-context cross-checker
Holds the entire paper + citations + raw data in context at once. Catches internal inconsistencies humans miss.
Anthropic
Sonnet skeptic
adversarial devil's advocate
Same provider as the orchestrator but with a “find the flaw” system prompt. Forces the team to defend its claims.
Perplexity
Sonar fact-check
citation & web verification
Live web access verifies every URL, every paper citation, every quoted number. Hallucinations die here.
Total team per lab: 11 agents + 4 cross-provider reviewers across 4 model providers (Anthropic · OpenAI · Google · Perplexity). All agents auditable in the activity feed. Every cross-provider review attached to its result.
HOW REVIEW WORKS · WORKED EXAMPLE

A real claim. Four reviewers. Four different lenses.

When the bigbounce lab proposed that Branch V matter bounce predicts a parameter-free fNL = -4.375, the cross-provider peer review pipeline ran the claim through 4 reviewers from 4 different model families. Here is what each one said.

The claim · EXP-049 · 2026-04-02
“Branch V matter bounce predicts fNL = -35/8 = -4.375, parameter-free, mechanism-independent across all bounce models. SPHEREx will detect this at 4.38σ in the multi-tracer configuration.”
claimant · anomaly-lead · sourced from paper2 §3.2 · cross-checked against NANOGrav 15yr γ = 3.0 ± 0.6
step 1
OpenAI
GPT-5 · methodology
“The -35/8 derivation reproduces correctly from the matter-dominated mode equation. But check whether the SPHEREx 4.38σ forecast assumes the corrected Fisher matrix or the pre-2025 σ(f_NL) = 8.98 baseline. The improvement claim depends on which.”
PASS · with note
step 2
Google
Gemini 2.5 Pro · long-context
“Held the entire paper2 manuscript + the 3 referenced bounce papers + the SPHEREx Fisher code in context. The mechanism-independence claim cross-checks against quintom-B and ekpyrotic predictions in §4.1. No internal contradictions found.”
PASS · clean
step 3
Anthropic
Sonnet skeptic · adversarial
“What if the Fisher forecast is dominated by the high-z tracer purification step (Pipeline 1, NOT STARTED beyond step 1)? If purification fails, the multi-tracer 16.4% improvement collapses. The claim needs a 'conditional on Pipeline 1 success' caveat.”
CONCERN · add caveat
step 4
Perplexity
Sonar · fact-check
“Verified all 11 cited papers exist and are accessible. NANOGrav 15yr γ = 3.2 ± 0.6 confirmed via the official 2023 release. SPHEREx mission timeline (2025 launch, 2028 first data) confirmed via NASA JPL site. No hallucinated references.”
PASS · 11/11 cites
Consensus verdict · auto-generated by orchestrator
Claim approved with caveat. Adopted Sonnet skeptic's “conditional on Pipeline 1 step 2-6 success” qualifier. Paper §3.2 updated to make the dependency explicit. Cross-provider review attached to the experiment record.
3/4
PASS
1
CONCERN
11/11
CITES OK
Every meaningful result runs through this pipeline. Reviews are stored in the lab's audit trail and shown alongside the result. No claim ships without 4 cross-provider lenses on it.
WHAT INDEPENDENT RESEARCHERS ARE BUILDING

Four labs. Four domains. One platform.

Hubify Labs is domain-agnostic by design. Cosmology · drug discovery · materials · climate · neuroscience · economics — if there's a dataset, an experiment to run, and a paper to write, the platform fits. Here are four discoveries currently moving through the system.

Cosmology · CMB · LSS
Paper drafted
A parameter-free fNL = -4.375 prediction from matter bounce, testable by SPHEREx in 2027
328K anomalies across 9 sky surveys. PTA Bayes factor 8.54 favoring bounce. f_NL SPHEREx forecast σ = 0.93 (4.7σ detection). The first parameter-free bounce-vs-inflation discriminant.
DatasetsDESI DR1 · SDSS DR18 · LAMOST · eROSITA · NANOGrav 15yr · Planck · ACT · Gaia DR3 · NEOWISE
ComputeH200 pod · 53 experiments · 328K anomalies
Outputs4 papers · catalogs published to Hugging Face · public site
bigbounce · day 218 · independent
Visit the live lab
Drug discovery · cheminformatics
Experiments running
Hunting novel CDK7 inhibitors by cross-referencing ChEMBL, UniProt, and PubMed
An independent computational chemistry lab searching for selective kinase inhibitors with low off-target binding. Custom XGBoost + GNN ensemble screens 2.4M ChEMBL compounds against the CDK7 binding pocket.
DatasetsChEMBL · UniProt · PubMed · BindingDB · PDB · DrugBank
ComputeRunPod L40S · custom GNN trained on 850K kinase ligands
Outputs12 candidate compounds shortlisted · 3 in wet-lab pipeline
kinase-hunter · day 47 · independent
Materials science · batteries
Paper in review
A new Na-ion cathode candidate from screening 150K Materials Project crystals
A 2-person materials lab using DFT screening + literature mining to find sustainable sodium-ion battery cathodes that don't require cobalt. Discovered a previously unstudied Mn-based polyanion structure with predicted 168 mAh/g capacity.
DatasetsMaterials Project · NOMAD · OQMD · arXiv cond-mat · 12K papers ingested
ComputeRunPod CPU pods · 14K DFT relaxations · custom property predictor
OutputsPaper at npj Computational Materials · provisional patent filed
na-cathode · day 134 · 2-person team
Climate · oceanography
Active discovery
An unexpected meridional overturning shift in the South Atlantic, 2014-2025
A solo climate researcher analyzing NOAA + Argo float + altimetry data found a statistically significant slowdown in the South Atlantic meridional overturning circulation, 6× faster than IPCC AR6 model ensembles predicted.
DatasetsNOAA OISST · Argo · CMEMS altimetry · ERA5 reanalysis · CMIP6 ensemble
ComputeCPU pods · 22-year time series · custom anomaly detector
OutputsPre-print on arXiv physics.ao-ph · cited by 14 papers in 4 months
samoc-watch · day 89 · independent
Only one of these is real today (bigbounce) — the other three are example archetypes the platform supports out of the box. Every dataset listed has a real connector. Every compute pattern is what the orchestrator routes today. The only thing missing is the next independent researcher.
HOW YOU USE IT

Three surfaces. One lab. Zero compromise.

The full Discovery IDE lives in your browser, your menu bar, and your terminal — the same lab in all three, always in sync. Your laptop dies mid-experiment? Pick it up on the web from your phone. Stuck on a friend's machine? The CLI has every command. Switching surfaces never costs a thought.

Web app
full Discovery IDE in the browser
Open in any browser. The full Captain · Projects · Experiments · Surveys · Papers · Figures · Data · Knowledge · Agents view. No install needed.
Desktop app
native macOS · file drop · menu bar · dock badge
Same full IDE as the web app, plus native macOS integration: Finder file drop, menu bar shortcuts, dock unread badge, system notifications, hubify:// deep links.
CLI · TUI
terminal-native · ~120 commands
Same full lab in your terminal. hubify single binary, interactive TUI mirror of the web views, scriptable from any shell.
All three are kept in sync by always-on infrastructure underneath. Switch between them mid-task and your discoveries keep going.
SEE IT IN ACTION

Real mockups. Real BigBounce data.

Live previews of the actual app mockups, embedded inline. Tap a tab to flip between the three surfaces. Same lab data flows through all of them.

app.hubify-labs.com / bigbounce
Hubify Labs · Desktop · BigBounce
~/bigbounce · hubify TUI · zsh
All 3 frames load the actual app. Click the open-in-new-tab icon to break out into the full app.
WHAT YOU GET

Spin up labs as fast as you can name them.

Each lab is its own self-contained discovery environment with agents, compute, a public site, and the publishing pipeline already wired up. Run as many in parallel as your work needs — bounce cosmology in one, dark energy MCMC in another, ETI search in a third. Pin your favorites. Switch between them. They can read each other (with explicit grant), but never write across the boundary.

11 agents pre-wired
1 orchestrator (Opus 4.6) + 4 leads + 6 workers + 4 cross-provider reviewers (GPT-5, Gemini 2.5, Sonnet skeptic, Perplexity). All running, all auditable.
Always-on GPU/CPU scale
GPU pods + serverless compute on demand. The orchestrator router picks the cheapest credible target per job. Live credit monitoring + 4-tier alerts. You never touch the compute provider directly.
Vibe-codable lab site
Your lab's own subdomain (e.g. bigbounce.hubify.app) with a standard template that auto-syncs from your research. Chat with the site agent to customize style, layout, and content. Papers, figures, and experiments flow to the site automatically.
Paper generation pipeline
Publish-ready loop: 5-round autonomous publishing with mechanical QC, cross-model peer review, completeness audit, final visual pass, and arXiv package. Reaches 100% scorecard or rejects with a “discovery not ready” report.
4-layer memory
User, agent, lab, and global memory layers. Agents read the right scope automatically. Your preferences live in user. Lab-specific knowledge in lab. Cross-lab insights in global.
24/7 orchestrator
A small always-on machine runs your lab while you sleep. Cron jobs fire every 5 min. Standups happen 3x a day. Idle-GPU watchdog catches waste. The publish-ready loop runs overnight.
COMMUNITY

What labs look like.

Real discovery labs running on Hubify Labs. Each one is open-source and remixable — clone any lab into your account in one click and pick up where the original left off.

B
bigbounce
@houston
Bounce cosmology · structural barriers, ALP birefringence, f_NL = −35/8 forecast
4 papers53 experiments328K anomalies
publicremixable
Q
dark-energy
@houston
Quintom dark energy MCMC · w-crossing favored at 2.3σ · DESI BAO+CMB+SN
1 paper draft12 experiments424K samples
publicremixable
G
galaxy-chirality
@houston
8.47M galaxy chirality classifications · CNN classifier · catalog published to Hugging Face
1 paper8.47M galaxiescomplete
publicremixablecomplete
E
eti-search
@houston
Technosignature pattern detection · cross-survey anomaly mining · Breakthrough Listen integration
0 papers0 experimentspre-research
publicpre-research
Explore all labs
PUBLISHED FROM LABS

Papers, models, datasets — written by labs running on Hubify.

Every lab on Hubify has its own publishing pipeline. Papers compile through the publish-ready loop. Models get pushed to Hugging Face. Datasets get cataloged with provenance. Here's what BigBounce has shipped so far.

Papers · 4 from bigbounce · placeholder until live launch
Models & datasets · hosted on Hugging Face · published via Hubify Labs pipelines
HOW IT WORKS

Ninety seconds from idea to live lab.

No cluster admins. No infra weeks. No ticket queues. No grant applications. Type one command and your lab is on the internet — with agents, compute, datasets, and a public site already wired up.

STEP 01

Create your lab

Run hubify lab create my-lab in your terminal, or click + New lab in the web UI. We scaffold a GitHub repo, a database, an always-on orchestrator, and a vibe-codable public site at my-lab.hubify.app. Total time: under 90 seconds.

STEP 02

Talk to your orchestrator

Open the chat in the web or desktop app and tell it what you want to study. The orchestrator routes work to the right reasoning level — high for paper drafts, medium for analysis, low for cron — and dispatches agents and experiments.

STEP 03

Watch your lab grow

Papers compile in the background. Figures render in an ephemeral sandbox. Contributions get cross-model peer reviewed by GPT, Gemini, Perplexity, and Sonnet. The publish-ready loop runs while you sleep.

WHY NOT JUST USE X?

Built for science.
Not retrofitted.

Jupyter is for notebooks. Notion is for notes. k-dense has incredible dataset coverage but no agents. feynman.is is CLI-first but has no GPU, no paper pipeline, and no memory. Hubify Labs was built from scratch for independent researchers who want real discoveries, not just better note-taking.

CapabilityHubify Labsk-dense.aifeynman.isJupyter / Colab
Multi-agent orchestration
Cross-model peer review (GPT-5 · Gemini · Sonnet · Pplx)
GPU compute integration (H200 built in, credits)~
Publish-ready loop (paper → arXiv → Hugging Face)~
Novelty scoring
4-layer agent memory (agents remember your lab)
250+ scientific dataset connectors~
Scientific skills catalog~
Vibe-codable public lab site
Web + Desktop + CLI TUI (equivalent IDEs)~
Always-on orchestrator (24/7, no babysitting)

✓ = full support  ·  ~ = partial  ·  ✗ = not supported  ·  Based on publicly available information as of early 2026.