Dyno
Dyno Phi

Calibrate and rank
your protein designs.

Skills for efficient agentic scoring, calibration and advancement of AI-generated protein designs without managing infrastructure.

01

Design

Generate binder candidates via the agent, the web app, or the REST API.

02

Score

Run scoring models on cloud GPUs to evaluate each design.

03

Calibrate

Shortlist and rank candidates against experimental calibrated filters.

04

Advance

Move shortlisted designs to the synthesis queue, share with collaborators, or hand off to the wet lab.

Phi Skills

One skill file gives your coding agent (e.g. Claude Code, Cursor) access to all Dyno Phi protein design capabilities through the phi CLI. Single commands wrap complex multi-step workflows, reducing token usage by ~90% on agentic design tasks.

Phiv1.0.0phi-skill.md

Single CLI tool for all Dyno Phi protein design capabilities — structure prediction, sequence design, language model scoring, and job management. One command to learn, all models accessible.

  • Structure prediction — ESMFold (fast screen) and AlphaFold2 (complex validation)
  • Sequence design — ProteinMPNN inverse folding with temperature control
  • Language model scoring — ESM-2 log-likelihood and perplexity
  • Biomolecular complex prediction — Boltz-1/2 for proteins, DNA, RNA
  • Biological research — PubMed, UniProt, PDB, and STRING queries with citations
  • Job management — submit, poll, cancel, download results

Setup

1 — Get an API key

# Get your key at design.dynotx.com/dashboard/settings
export DYNO_API_KEY="your_key_here"

2 — Add the skill to your agent

# In Claude Code, run the plugin install command:
/plugin install dyno-phi

The Dyno Phi skill is listed in the Claude Code Skills Marketplace. Run /plugin install dyno-phi in any Claude Code session to add it — no file management needed.

Install the CLI manually

# Requires Python ≥ 3.11
uv tool install dyno-phi

New to uv? docs.astral.sh/uv — a fast Python package and project manager. Use uv tool install to make phi available globally without activating a virtualenv.

Core workflow

Research

# Understand the target before committing to a design campaign
phi research --question "What are the key binding hotspots on PD-L1?"

Design

# Generate binder candidates against a target structure
phi design --target target.pdb --num-designs 10000 --out ./candidates

Filter

# Score, calibrate, and rank — apply default or relaxed thresholds
phi filter --preset default --wait

More commands

  • phi esmfold — fast single-sequence structure screen
  • phi alphafold — high-accuracy complex validation
  • phi esm2 — log-likelihood and perplexity scoring
  • phi upload — bulk-upload FASTA/PDB datasets
  • phi scores — view per-design scoring results
  • phi jobs — list, poll, and cancel running jobs

All compute runs on Dyno cloud GPUs — no local GPU required. Full CLI reference →