The toolbox · 43 tools and 14 datasets
What to use, by what you’re trying to do.
Each entry says what it does, how to get it, and how people use it, with the source.
Read and check papers
Evidence API
Searches 100M+ papers for the strongest evidence for and against a claim.
How people use itJames Zou's group built it and shows it on one question: does drinking coffee reduce diabetes risk?
Try it (our suggestion)Paste a claim from your grant draft and read what argues against it.
Paper Scorecard
Finds the citing papers that actually engage with a paper and shows their critiques.
How people use itNo documented use yet.
Try it (our suggestion)Run it on a paper before your lab builds on it.
Consensus, now with Science journals
Consensus can now search and cite the full text of every article in the Science family.
How people use itAAAS put the full text of the Science journals into Consensus. It reports clicks per search appearance more than doubled since August (AAAS's own figure). Proof Source is a trade-news repost of the AAAS announcement.
QED Science
Uses AI to judge published claims on their merits, apart from the journal they ran in.
How people use itQED Science scored 57,455 bioRxiv preprints and published a top-574 list. Nature asked whether researchers should trust the scores. Proof The scores and the list are QED's own claims. No independent check found.
Blossom
A database of psychedelic research: paper summaries, trial data and a results explorer.
How people use itNo documented use yet.
onco.cc
An open map of oncology: cancers, products, technologies, bottlenecks and people.
How people use itJude Gomila built it as a free resource and asks technical readers to file fixes in the repo.
Analyze your lab data
Cytofeather
A lightweight flow cytometry app that runs in your browser.
How people use itJonas Koeppel built it with Codex and Claude Code as a free alternative to paid biology software.
A lab's own flow cytometry app
An immunology lab building its own flow cytometry analysis software with GPT-6 Astra.
How people use itDerya Unutmaz (@DeryaTR_) is building it because labs like his pay thousands of dollars a year in commercial license fees.
Lazarus
An agent that gets dead research code running again, starting from a GitHub link.
How people use itDean Sherry sampled 20 bioinformatics tools from 2018 to 2021. 17 no longer installed or ran. Lazarus won the Build with Claude Life Science Hackathon.
Try it (our suggestion)Point it at the old repo your methods depend on.
bioclaw
An open agent harness for running biological foundation models.
How people use itDarren Mckeeman built it while prototyping day-to-day tools with AI. Elliot Roth plans to use it for transcriptomics side projects.
AutoScientist
Automates model training from an objective you set, then exports to Hugging Face.
How people use itNo documented use yet. At launch the company offered a free 30-day trial. Current terms not confirmed.
ScienceBuddy
A free workspace for scientific agents, with GPT-6 access at no cost.
How people use itNo documented use yet. The site is in Chinese and shows no pricing.
DeerFlow (ByteDance)
An open-source agent that splits a long task across sub-agents in its own sandbox. General purpose, not built for science.
How people use itNo documented use yet.
Resipher
Measures mitochondrial respiration for days in standard 96-well plates, inside the incubator. A lab instrument; the post does not describe AI.
How people use itNo documented use yet.
WeatherNext 3
Google's global weather model, with hourly high-resolution forecasts anywhere on Earth through Weather Lab.
How people use itNo documented use yet. Google's launch post lists BigQuery, Earth Engine and Cloud Storage for the data. A separate Weather Lab was not found.
Find data to reuse
Neuroscience dataset browser
Search 10,000+ open neuroscience datasets across EEG, MRI, MEG, NIRS, PET and more, from about 1,000 institutions.
How people use itNo documented use yet. MIT Media Lab lists it as NeuroDatasets.
Try it (our suggestion)Search your modality and task before you collect new data.
The full arXiv corpusDataset
All 3,148,796 arXiv papers, every version, 16 TB, as a Hugging Face dataset.
How people use itNo documented use yet.
OpenAlex snapshot
The full OpenAlex record of scholarly work as one free download, with new machine-readable docs.
How people use itNo documented use yet.
Try it (our suggestion)Download it once instead of scraping the website.
Digital Brain ProjectDataset
Planned: 15,000 hours of fMRI, MEG and intracranial recordings from 100 people at 14 centers. Meta committed $15M.
How people use itNine funded teams, including Jorn Diedrichsen at Western, Daphne Bavelier at Geneva and Nikolaus Kriegeskorte at Columbia, will record brain activity while people play games, for one shared open dataset. Proof Planned. No data released yet.
OpenMHCDataset
More than a decade of My Heart Counts wearable data, 11,894 participants.
How people use itNo documented use yet.
MAESTRO dataDataset
A researcher describes it as open, dense and high quality. What it contains is still to confirm.
How people use itExample to source.
ChronicleBioDataset
A company building a deep dataset on POTS, Long COVID, ME/CFS and hEDS, to match drugs to patients.
How people use itNo documented use yet. Forbes piece not retrieved. Partnership with Solve M.E. and others reported in a press release.
Synthetic clinical conversationsDataset
Synthetic clinical conversations and retrieval cards for every human disease, generated by a model. Synthetic, so check it before you train on it.
How people use itNo documented use yet.
Design proteins and molecules
Codex with the Adaptyv plugin
Design peptides in Codex, then send them to Adaptyv's lab for testing.
How people use it@theo_jala's four steps: join the Rosalind Workbench preview, design with a model such as Boltz, send the designs to Adaptyv, then design again from the lab data.
Rosalind Workbench
A workbench for biologists, built into Codex.
How people use itNo documented use yet.
bio.rodeo
Tracks 1,796 biological foundation models, with alerts by category and an MCP server.
How people use itJustin Kiggins built it after 203 new bio models shipped in 90 days, so he could stop checking by hand.
Try it (our suggestion)Follow the one category you work on and take the weekly digest.
JapanFold
A platform for serving bio models, usable from Claude, Cursor, Codex or a web app. Its cost claims are its own.
How people use itNo documented use yet. Launch coverage says the browser workbench needs no install or sign-up. Cost claims are the company's own.
MagNET on Rowan
An ML model that predicts 1H and 13C NMR shieldings, a drop-in for DFT in NMR workflows. Built with groups at Merck, MIT, Harvard and CSU.
How people use itCorin Wagen's Rowan runs it as a hosted NMR predictor: draw a structure, pick a solvent and get 1H and 13C shifts in seconds. It was built with Eugene Kwan's group at Merck and groups at MIT, Harvard and Colorado State. Proof
Check a model before you trust it
Terminal-Bench-ScienceDataset
70 research-workflow tasks for AI agents across fields. Claude Opus 5 solved about 30%.
How people use itA Stanford-led team with working scientists ran AI agents on 70 tasks taken from researchers' own workflows. At launch the best, Claude Opus 5, solved about 30%. Proof Launch figure for v0.1. Public leaderboards now list higher scores for newer models, so check the leaderboard before quoting a number.
Try it (our suggestion)Ask any agent vendor for its score on a benchmark like this one.
BixBench3Dataset
Can a model reproduce the analyses behind a biology paper from raw data? The best of 13 models got 48% of the artifacts.
How people use itEdison Scientific gave 13 frontier models the raw data from 20 biology papers and asked them to redo the analyses. The best reproduced 48% of the requested results. Proof
Antibody Discovery BenchmarkDataset
100 evaluations across ten stages of antibody discovery, drawn from real drug programs.
How people use itNo documented use yet.
OncologyBenchDataset
25 multistep oncology evaluations, built on 92 tasks, across six stages of cancer care.
How people use itLatchBio tested 31 model-and-harness combinations on 25 cancer decisions. The best, Claude Opus 5 on Pi, passed 40% of attempts. Proof
Bedside BenchDataset
Doximity's clinical AI benchmark: 500 questions written by physicians, across 10 parts of clinical care.
How people use itFireworks AI graded models on Doximity's 500 physician-written cases as part of its Specialized Intelligence Index. Doximity reports its own Doximity Ask ranked first. Proof The first-place result is Doximity's own report.
AutoMedBenchDataset
Tests medical AI agents at every step (planning, setup, validation) as well as the final answer. 6,300+ runs, 7 tracks.
How people use itUC Santa Cruz and NVIDIA scored medical research agents on planning, setup, validation, inference and submission across 6,300+ runs. Validation was the weakest stage. Proof
MentalHealthBenchDataset
An open benchmark, built with mental health experts, for how AI responds when people are struggling.
How people use itOpenAI's Karan Singhal and colleagues scored models on realistic mental health conversations graded by clinicians. Models differed widely at asking for context and were more alike on empathy. Proof
A biopharma task benchmarkDataset
@andrewwhite01 says it mimics real biopharma tasks well. Its name is still to confirm from the source.
How people use itExample to source.
Clinical and preclinical work
whydidthistrialfail.ai
Enter a trial ID, sponsor or title and get a structured explanation of why the trial likely failed.
How people use itMahesh Narasimhan built it by joining ClinicalTrials.gov records with PubMed and other public evidence.
Try it (our suggestion)Run it on a failed trial in your target area before you design the next one.
MedDecider
Open-weight decision models for medicine. The team reports its 9B model beat a model three times larger.
How people use itNo documented use yet. Model card lists a non-commercial license (CC BY-NC 4.0) and a lower score on unseen medical exams than the team's headline suggests.
MONAI Physio
An open-source toolkit that turns 3D and 4D medical images into heart and lung digital twins.
How people use itNo documented use yet.
NAMina Program Builder
Designs human-relevant preclinical programs around the decision the program has to make.
How people use itPaola Dama ran it back over a published pancreatic cancer tumoroid study and scored seven design areas. She says one self-scored case shows it can represent a program, not that it works in general. Proof
Genome Intelligence
Explore a genome with AI models without handing the models the genome. A consumer product; useful as a privacy pattern.
How people use itNo documented use yet.
Personal health baselines
A Cell Systems paper from Leroy Hood's group: measure each person against their own baseline instead of a population average.
How people use itNo documented use yet.
Find funding
ResearchHub
Matches researchers to open RFPs, reads their published work and drafts a preregistered proposal.
How people use itNo documented use yet. Search found no docs confirming the automatic matching and drafting pipeline the post describes.
UK research grants database
200,000+ UK research projects, almost 100bn pounds of funding, with semantic search.
How people use itNo documented use yet. The site lists about 270,000 records, more than the post's 200,000.
Try it (our suggestion)See who funded work like yours before you apply.
GrantSource
Finds funders and drafts grant proposals for nonprofits.
How people use itNo documented use yet. Started as a team project at an Emory event. No public site found.
TechBio 50
A list of the most active AI and biology venture investors, from Atria.
How people use itNo documented use yet.
29 VC funds and their focus
A plain list of venture funds and what each one backs, including science and deep tech funds.
How people use itNo documented use yet.
Explain and share your work
PaperBanana (Google)
Turns methodology text into figures for papers.
How people use itNo documented use yet. Google's version is a research paper (arXiv 2601.23265). The free tools are unofficial community builds that need your own image-model API key.
Try it (our suggestion)Paste a methods section and edit the draft diagram it gives you.
arXivisual
Add "isual" to an arXiv link and get an animated, narrated walk through the paper.
How people use it@hasantoxr tried it on the original Transformers paper and says attention finally made sense.
ProteinBlender
Free, open-source molecular animation, now with a DNA builder and short tutorials.
How people use itNo documented use yet.
Color-coded equations
Color-codes an equation and links each part to a plain explanation of its role.
How people use itNo documented use yet.
iPlaces
Publishing built around persistent IDs (ORCID, ROR, DOIs), so field stations can show the research they support.
How people use itNo documented use yet. 'Live in production' is a status, not an access term. Project records describe a staging demo and say the team is stabilizing the service.
Get access and learn
Anthropic Life Sciences Verification Program
Verified labs and companies can apply for access to Anthropic's most capable models for biology and drug development work.
How people use itNo documented use yet.
Try it (our suggestion)If Claude blocks your biology work, apply here.
BioTorch
A learning path into biological AI research, in public beta.
How people use itNo documented use yet.
Google Foresight app
A free, experimental Google app for Mac that turns meeting shorthand into full notes, offline.
How people use itNo documented use yet. Search coverage describes a free, experimental Mac meeting-notes app that runs offline. Not a science tool as far as found.
About this list
21 of 55 entries show a named person or lab using the tool, with a link. 32 are announcements only and say “No documented use yet.” 2 are still being checked. “Try it” lines are our suggestions, not reports of real use. Access marked “unchecked” comes from the post, not the tool’s own page. Updated October 26, 2026.
Know a real use for a tool here? Tell me and I’ll add it with your name.