Steinegger Lab
Proto
Google DeepMind
ByteDance
Biohub
Dunbrack Lab
IPD
Google DeepMind
Steinegger Lab
UT Southwestern 
St. Jude
Broad Institute 
The Jackson Laboratory 
Yale University
RIMD
Google DeepMind
Steinegger Lab 
Söding Lab
Microsoft Research
IPD
Biohub
RCSB
Google DeepMind
UNR UW 
UQ 
UCSD
EMBL-EBI
Calico
Google DeepMind
Rutgers University 
Columbia University
Google DeepMind EMBL-EBI
Biohub
Biohub
Meta AI Biohub
SIB EMBL-EBI 
PIR
EMBL-EBI
PDBe EMBL-EBI 
wwPDB
Aureka AI Research
Arc Institute
Chai Discovery
Salesforce Research
MSKCC
UW
Proto
Arc Institute
Meta AI
NCBI
NCBI
IPD
Arc Institute
NCBI
NCBI
Meta AI Biohub
IPD
Profluent
TBI Vienna
UCSF 
Autosome.org
RosettaCommons
Boltz 
MIT Jameel Clinic 
Recursion
USC
OPIG Steinegger Lab Proto Google DeepMind ByteDance BiohubDunbrack Lab
IPD Google DeepMind Steinegger Lab UT Southwestern St. Jude Broad Institute The Jackson Laboratory Yale University RIMD Google DeepMind Steinegger Lab Söding Lab Microsoft Research IPD Biohub RCSB Google DeepMind UNR UW UQ UCSD EMBL-EBI Calico Google DeepMind Rutgers University Columbia University Google DeepMind EMBL-EBI Biohub Biohub Meta AI Biohub SIB EMBL-EBI PIR EMBL-EBI PDBe EMBL-EBI wwPDB Aureka AI Research Arc Institute Chai Discovery Salesforce Research MSKCCUW
Proto Arc Institute Meta AI NCBI NCBI IPD Arc Institute NCBI NCBI Meta AI Biohub IPD Profluent TBI Vienna UCSF Autosome.org RosettaCommons Boltz MIT Jameel Clinic Recursion USC OPIGThe Open-Source Infrastructure Layer for Biology
bash
Proto Tools
A lot of computational biology looks like this: find a tool that does what you need, spend half a day getting it to install, spend another day figuring out how to call it, then reformat the output so the next tool can use it. Every tool has its own opinions about Python versions, CUDA, weight paths, and input formats.Proto Tools implements these tools so they all behave the same way. Each one is a Python function with a Pydantic input and output. Fold a protein with AlphaFold3, sample sequences with ProteinMPNN, search with BLAST; the shape of the call is the same, and the first invocation quietly sets up the environment and weights in isolation.Use it on its own, or as the tools layer of the broader Proto framework.Key Features
A quick tour of what proto-tools handles for you. Each card opens a deeper guide.Unified Interface
Consistent Python APIs with standardized Pydantic Input/Config/Output models across all tools
Isolated Environments
Each tool gets its own virtual environment; weights and dependencies are fetched and cached on first use
Device Management
Automatic GPU allocation, LRU eviction, and multi-GPU routing
Parallel Execution
Fan a single tool call out across every visible GPU with ToolPool
Tool Catalog
Every tool we implement, at a glance. Select any to open its page.Get Started
Installation
Set up the package and optional dependencies
Quickstart
Run your first tool in 5 minutes
Concepts
Understand the tool pattern and execution model