generated: '2026-08-10'
method: searched
source: https://design.dynotx.com/cli
description: >-
  `phi` is Dyno's first-party command-line interface for the Phi Protein Design
  API — the primary way the platform is used, and the surface the provider's own
  Claude Code Agent Skill drives. It wraps the REST API at
  https://api.dyno-agents.app/v1/phi. The distributable package itself is
  catalogued in packages/dyno-packages.yml; this artifact enumerates the command
  surface.
docs: https://design.dynotx.com/cli
repo: https://github.com/dynotx/phi-cli
license: MIT
binary: phi
package: dyno-phi
version: 0.2.3
requires: Python >= 3.9
install:
  pip: pip install dyno-phi
  uv: uv tool install dyno-phi
authentication:
  env_var: DYNO_API_KEY
  key_prefix: ak_
  issue_at: https://design.dynotx.com/dashboard/settings
  verify: phi login
  cache: .phi/state.json (api_key, dataset_id, last_job_id, user_id, org_id)
  base_url_override: DYNO_API_BASE_URL (default https://api.dyno-agents.app)
commands:
  auth:
  - {name: login, description: Verify the API key and connection to the platform.}
  data:
  - {name: upload, description: Upload PDB/CIF files or a whole directory as a new dataset.}
  - {name: fetch, description: Fetch a structure from RCSB PDB or the AlphaFold DB; optionally crop by chain or residue range and trim by pLDDT, then upload.}
  - {name: use, description: Set the active dataset, cached to .phi/state.json.}
  - {name: datasets, description: List all datasets.}
  - {name: dataset, description: Show details for one dataset.}
  - {name: ingest-session, description: Multi-file ingest session for large uploads.}
  modeling:
  - {name: folding, alias: esmfold, description: Single-sequence structure prediction (ESMFold).}
  - {name: complex_folding, alias: alphafold, description: Multi-chain complex prediction (AlphaFold2 multimer).}
  - {name: inverse_folding, alias: proteinmpnn, description: Sequence design from a backbone (ProteinMPNN).}
  - {name: boltz, description: Structure prediction with Boltz-1.}
  - {name: esm2, description: ESM2 sequence embedding and scoring.}
  scoring:
  - {name: filter, description: "Full quality-control pipeline: inverse folding -> folding -> complex folding -> score, against a threshold preset."}
  - {name: scores, description: Display the scores table for a completed filter job; --top N, --out CSV.}
  jobs:
  - {name: status, description: Poll the status of one job.}
  - {name: jobs, description: List recent jobs; --status filter.}
  - {name: logs, description: Stream (or print the URL for) a job's log stream.}
  - {name: cancel, description: Cancel a running job.}
  - {name: download, description: "Download job artifacts into structures/, scores/ and scores.csv; --all includes MSA files and archives."}
  research:
  - {name: research, description: Run a biological research query with literature citations against the platform, optionally streaming and appending to a notes file.}
  - {name: notes, description: Fetch and display accumulated research notes for a dataset.}
  - {name: tutorial, description: Download five example PD-L1 binder structures and print a step-by-step scoring walkthrough.}
  feature_flagged:
  - {name: design, alias: rfdiffusion3, description: Backbone diffusion. Hidden from --help since 0.2.2; reachable via DYNO_ENABLE_DESIGN=1.}
  - {name: boltzgen, description: All-atom binder design. Hidden from --help since 0.2.2; reachable via DYNO_ENABLE_DESIGN=1.}
common_flags:
- {flag: --dataset-id ID, applies_to: most, description: Target dataset; omit to use the cached one.}
- {flag: --wait, applies_to: most, description: Poll until the job completes, then print a summary.}
- {flag: --out DIR, applies_to: [download, scores], description: 'Output directory (default ./results).'}
- {flag: --preset default|relaxed, applies_to: filter, description: Filter threshold preset.}
- {flag: --num-sequences N, applies_to: inverse_folding, description: 'Sequences per design (default 4).'}
- {flag: --models 1,2, applies_to: complex_folding, description: 'AlphaFold2 model numbers (default 1,2,3).'}
- {flag: --msa-tool mmseqs2|single_sequence, applies_to: [filter, complex_folding], description: MSA strategy; single_sequence for de novo binders with no natural homologs.}
- {flag: --poll-interval S, applies_to: global, description: 'Seconds between status polls (default 5).'}
filter_presets:
  default: {plddt: '>=0.80', ptm: '>=0.55', iptm: '>=0.50', ipae: '<=10.85 A', rmsd: '<=3.5 A', msa_tool: single_sequence}
  relaxed: {plddt: '>=0.80', ptm: '>=0.45', iptm: '>=0.50', ipae: '<=12.4 A', rmsd: '<=4.5 A', msa_tool: single_sequence}
  source: github.com/dynotx/phi-cli src/phi/config.py::_FILTER_PRESETS
  note: >-
    The published docs page and SKILL.md state iPAE as a normalized 0.35/0.40;
    the shipped code uses 10.85/12.4 Angstrom (0.35 x 31 per the BindCraft
    convention). Values captured verbatim from the code, which is authoritative.
key_flows:
- name: research-guided scoring
  steps: [phi research --question ... --target ... --structures, phi fetch --pdb ... --chain A --upload, phi filter --preset relaxed --wait, phi scores]
- name: score an existing design set
  steps: [phi upload ./designs/, phi filter --preset default --wait --out ./results, phi scores]
- name: first run
  steps: [pip install dyno-phi, export DYNO_API_KEY=..., phi login, phi tutorial, phi filter]
agent_integration:
  claude_code_skill: skills/dyno-phi.md
  claude_code_plugin: https://github.com/dynotx/phi-cli/tree/main/phi-plugin
  marketplace_manifest: .claude-plugin/marketplace.json
  allowed_tools: Bash(phi *)
x-evidence:
  fetched: '2026-08-10'
  urls:
  - {url: 'https://design.dynotx.com/cli', http_status: 200}
  - {url: 'https://raw.githubusercontent.com/dynotx/phi-cli/main/README.md', http_status: 200}
  - {url: 'https://raw.githubusercontent.com/dynotx/phi-cli/main/src/phi/config.py', http_status: 200}