dyno-phi

Retired. dyno-phi 0.2.x is the earlier Phi client and 0.2.4 is its last release; the API it talks to has been shut down, and every 0.2.x release on PyPI is yanked. The current Phi client is dyno-phi 1.0 and later, published from dynotx/phi-api under the same package name: run pip install --upgrade dyno-phi (or uv tool install "dyno-phi[cli]"; Python 3.12+), set PHI_API_KEY and PHI_API in place of DYNO_API_KEY and DYNO_API_BASE_URL, and read the new command tree at design.dynotx.com/docs. This repository is archived. The text below describes 0.2.x as it was.

Phi CLI for the dyno protein structure analysis platform.

phi is the command-line interface for uploading protein structures, running structure prediction and inverse-folding pipelines, filtering and scoring candidates, and downloading results from the dyno API.

Results and scores are viewable in the web dashboard at design.dynotx.com/dashboard/datasets/<dataset_id>.

Table of contents

• Installation
• Authentication
• Quick start
• Command reference
• Filter presets
• State caching
• Claude Code skill
• Development

Installation

pip install dyno-phi

Requires Python ≥ 3.9.

Authentication

Create an API key at Settings → API keys in the dyno web app (https://design.dynotx.com/dashboard/settings), then export it:

export DYNO_API_KEY=ak_...

The key is cached to .phi/state.json on first use so you don't need to re-export it in future sessions.

Verify your connection:

phi login

Quick start

Try the tutorial

The fastest way to get started — downloads five example PD-L1 binder structures and walks you through the full pipeline:

phi tutorial

This fetches the example files, prints step-by-step instructions, and leaves you ready to run phi filter.

Scoring your own structures

# 1. Upload PDB/CIF files
phi upload ./designs/

# 2. Run the full filter pipeline
phi filter --preset default --wait

# 3. View scores and download results
phi scores
phi download --out ./results/

After each command, phi prints the active dataset and a link to the dashboard:

Active: dataset [d7c3a1b2-...] · job [cb4553f5-...]
Dashboard: https://design.dynotx.com/dashboard/datasets/d7c3a1b2-...

Command reference

Common flags

Filter presets

phi filter applies a multi-stage quality-control pipeline and scores each design against configurable thresholds.

Override any threshold with an explicit flag:

phi filter --dataset-id ... --plddt 0.75 --iptm 0.45

State caching

phi caches the most recently used dataset ID, job ID, and API key in .phi/state.json so you don't need to pass --dataset-id or re-export your key repeatedly:

phi use d7c3a1b2-...         # set active dataset
phi filter --preset default  # uses cached dataset
phi scores                   # uses cached job
phi download --out ./results # uses cached job

The dashboard URL for the active dataset is printed after every command:

Dashboard: https://design.dynotx.com/dashboard/datasets/d7c3a1b2-...

Claude Code skill

The phi skill is bundled at skills/phi/SKILL.md and is automatically available when you open this repo in Claude Code (Cursor or the claude CLI). No installation needed — just open the project and ask naturally:

Upload the PDB files in ./examples/ and run the default filter pipeline.

To make the skill available in all your projects (outside this repo):

mkdir -p ~/.claude/skills/phi
cp skills/phi/SKILL.md ~/.claude/skills/phi/SKILL.md

Development

git clone https://github.com/dynotx/phi-cli
cd phi-cli
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"

# Lint
ruff check src/

# Type check
mypy src/phi/

# Run tests
pytest tests/

Releasing to PyPI

Releases are published via GitHub Actions. Push a version tag to trigger the workflow:

# Bump src/phi/_version.py, update CHANGELOG.md, then:
git tag v0.1.0
git push origin main --tags

The workflow publishes to TestPyPI automatically, then waits for manual approval before publishing to the real PyPI. See .github/workflows/publish.yml for details.