Setup

Setup Guide

Help users get their environment ready to run protein design tools.

Quick checklist

Run through this checklist when a user encounters setup issues:

| Step | Check | Fix | |------|-------|-----| | 1. Modal CLI | modal --version | pip install modal | | 2. Modal auth | modal token show | modal setup | | 3. biomodals | ls biomodals/modal_*.py | git clone https://github.com/hgbrian/biomodals | | 4. Test | cd biomodals && modal run modal_boltzgen.py --help | See troubleshooting |

Diagnosing issues

Error: "modal: command not found"

Cause: Modal CLI not installed.

Fix:

pip install modal

Then restart the terminal or run hash -r.

Error: "Permission denied" or "Unauthorized"

Cause: Modal not authenticated.

Fix:

modal setup

This opens a browser. Click "Authorize" to complete authentication.

Error: "No such file or directory: modal_boltzgen.py"

Cause: biomodals repository not cloned or not in correct directory.

Fix:

git clone https://github.com/hgbrian/biomodals
cd biomodals

Error: "uvx: command not found"

Cause: uvx is an optional wrapper from the uv package. It's not required.

Fix: Run modal directly (recommended):

modal run modal_boltzgen.py --help

Or install uv if you prefer using uvx:

pip install uv

Full setup steps

Step 1: Install Modal CLI

pip install modal

Verify: modal --version

Step 2: Authenticate Modal

modal setup

This opens a browser. Click "Authorize".

Verify: modal token show

Step 3: Clone biomodals

git clone https://github.com/hgbrian/biomodals
cd biomodals

Verify: ls modal_*.py should show files like modal_boltzgen.py

Step 4: Test the Setup

cd biomodals
modal run modal_boltzgen.py --help

Expected: Usage instructions appear showing --input-yaml, --protocol, --num-designs options.

Common workflows after setup

Once setup is complete, users can:

cd biomodals

# Design binders with BoltzGen (requires YAML config)
modal run modal_boltzgen.py --input-yaml binder.yaml --protocol protein-anything --num-designs 50

# Generate backbones with RFdiffusion (official repo, not biomodals)
python run_inference.py inference.input_pdb=target.pdb contigmap.contigs=[A1-150/0 70-100] inference.num_designs=100

# Validate with Chai
modal run modal_chai1.py --input-faa designs.fasta

GPU selection

Set GPU with environment variable:

GPU=L40S modal run modal_boltzgen.py --input-yaml config.yaml --num-designs 50
GPU=A100 modal run modal_chai1.py --input-faa complex.fasta

| GPU | VRAM | Best For | |-----|------|----------| | T4 | 16GB | ProteinMPNN, ESM | | A10G | 24GB | RFdiffusion, Chai | | L40S | 48GB | BoltzGen, BindCraft | | A100 | 40-80GB | Large complexes |

Modal free tier

Modal offers $30/month in free credits - enough for:

• ~500 BoltzGen designs
• ~2000 RFdiffusion backbones
• ~1000 Chai predictions

---

Full documentation: See [Getting started](../../docs/getting-started.md)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: adaptyvbio
• Source: adaptyvbio/protein-design-skills
• License: MIT

Install and usage instructions live in the source repository linked above.