tt-bio bundles third-party code under tt_bio/_vendor/ and tt_bio/boltzgen/.
Each component keeps its upstream license alongside it. The core Boltz-2 / Boltz-1
code that tt-bio is built on (parsing, tokenization, featurization, the model) is
MIT-licensed and credited in the top-level LICENSE.

1. tt_bio/_vendor/esm/ — ESMFold2 host-side reference: sequence/atom
   featurization, MSA handling, and mmCIF assembly (no neural compute; that runs
   on Tenstorrent via tt_bio.esmc / tt_bio.esmfold2).
   Source:        github.com/Biohub/esm @ b6b0e88
   License:       MIT — Copyright 2026 Chan Zuckerberg Biohub, Inc.
                  (tt_bio/_vendor/esm/LICENSE)
   Modifications: only the files on the ESMFold2 fold path are included; absolute
                  `esm.` imports rewritten to `tt_bio._vendor.esm.`.

2. tt_bio/_vendor/esmfold2_hf/ — the ESMFold2 model definition for HuggingFace
   `transformers` (runs on top of the stock `transformers` wheel).
   Source:        Biohub `transformers` fork (based on transformers 4.57.6),
                  src/transformers/models/esmfold2
   License:       Apache-2.0 — Copyright The HuggingFace team and Chan Zuckerberg
                  Biohub, Inc. (tt_bio/_vendor/esmfold2_hf/LICENSE)
   Modifications: the three inference files only; transformers-core relative
                  imports rewritten to absolute `transformers.`.

3. tt_bio/boltzgen/ — BoltzGen binder-design source, stripped to Tenstorrent-only
   inference (training code, PyTorch Lightning, and Hydra removed); the heavy
   compute runs on Tenstorrent via tt_bio.boltzgen.adapter.
   Source:        github.com/HannesStark/boltzgen
   License:       MIT — Copyright (c) 2025 Hannes Stärk (tt_bio/boltzgen/LICENSE).
                  Some modules carry their own upstream headers (lucidrains/
                  alphafold3-pytorch, MIT; PyTorch3D, BSD) — kept in-file.
   Modifications: inference-only subset; Lightning/Hydra dependencies eliminated.

4. tt_bio/protenix.py, protenix_data.py, protenix_weights.py — Protenix-v2
   (AlphaFold3-family) support. This is an INDEPENDENT ttnn reimplementation for
   Tenstorrent; no upstream Protenix source is vendored. The architecture is from
   ByteDance's Protenix. tt-bio neither redistributes nor downloads the Protenix-v2
   weights: upstream's 2026-04-08 Protenix-v2 release note says they "are not
   released under any open-source license, and may not be reproduced, distributed,
   sublicensed, disclosed, or otherwise transferred to any third party in any form
   without the express prior written consent of the rights holder." The protenix-v1
   checkpoint (upstream's v0.5.0, model_v0.5.0.pt) downloads on first use.
   Source:        github.com/bytedance/Protenix
   License:       Apache-2.0 — Copyright ByteDance (code; the v0.5.0 parameters).
                  Protenix-v2 parameters: proprietary, per the note above.

5. tt_bio/_vendor/openfold3/ — OpenFold3 host-side data pipeline: the query
   schema, CCD/ligand lookup, tokenization, structure / reference-conformer /
   MSA / template featurization, and the Lightning Dataset/DataModule training
   framework with its LMDB dataset caches (no neural compute; that runs on
   Tenstorrent via tt_bio.openfold3*).
   Source:        github.com/aqlaboratory/openfold-3, PyPI `openfold3` 0.4.3
   License:       Apache-2.0 — Copyright 2026 AlQuraishi Laboratory
                  (tt_bio/_vendor/openfold3/LICENSE)
   Modifications: core/utils/relpos.py is taken from 0.4.5, which added the
                  cyclic-offset helpers, and create_paired_from_precomputed
                  from 0.5.0 (upstream PR #373: 0.4.3 dropped precomputed
                  paired MSA rows); everything else is 0.4.3. Imports are
                  rewritten to the tt_bio._vendor namespace. 15 modules differ
                  from upstream: seven are import plumbing, eight carry
                  behavioural changes that are flag-gated and default to
                  upstream behaviour (see docs/openfold3-vendor.md).
                  memory_profiler, lmdb, boto3 and kalign are imported lazily
                  so inference does not require the training-only deps.
                  scripts/of3_port/vendor_of3_training.py regenerates the training
                  half and verifies it with --check.

6. tt_bio/openfold3*.py — OpenFold3 (AlphaFold3-family) on-device model. Like
   Protenix-v2 this is an INDEPENDENT ttnn reimplementation; no upstream compute
   code is vendored (only the data pipeline above). The weights
   (of3-p2-155k.pt, the OpenFold3 preview2 checkpoint) are the OpenFold
   consortium's ungated public parameter release; tt-bio does not redistribute
   them and downloads them from upstream's bucket on first use (OF3_CKPT or a
   file at ~/.boltz/ is used instead when present). OpenBind-0's
   of3-ob-2025-06-30-174k.pt comes from the same bucket the same way.
   Source:        github.com/aqlaboratory/openfold-3, weights from
                  s3://openfold3-data/openfold3-parameters/
   License:       Apache-2.0 — Copyright 2026 AlQuraishi Laboratory, code and
                  parameters ("The OpenFold3-preview model is released under
                  Apache 2.0 license", huggingface.co/OpenFold/OpenFold3).

7. tt_bio/_vendor/atomworks/ — AtomWorks: the AF3-style structure-processing and
   transform pipeline RF3 featurizes with (CCD lookup, bond perception,
   atomization, chirality perception, MSA pairing, templating). Host-side only.
   Source:        github.com/RosettaCommons/atomworks, PyPI `atomworks` 2.2.1
   License:       BSD-3-Clause — Copyright (c) 2025, Institute for Protein Design,
                  University of Washington (tt_bio/_vendor/atomworks/LICENSE)
   Modifications: absolute imports repointed at the vendor namespace, and two
                  Python 3.11-only imports replaced so the package runs on 3.10,
                  which is tt-bio's deployed runtime: `enum.StrEnum` via
                  tt_bio/_vendor/atomworks/_compat.py, and `typing.Never` via
                  typing_extensions. Verified bit-exact against the upstream 3.12
                  behaviour; see scripts/rf3_port/.

8. tt_bio/_vendor/rf3/ and tt_bio/_vendor/foundry/ — the RF3-specific host
   transforms (ground-truth templating, cyclic chains, the AF3 pipeline
   definition) and three helpers they call. No model or training code.
   Source:        github.com/RosettaCommons/foundry, models/rf3 and src/foundry
   License:       BSD-3-Clause — Copyright (c) 2025, Institute for Protein Design,
                  University of Washington (tt_bio/_vendor/rf3/LICENSE)
   Modifications: only the featurization subset is included. `foundry/__init__.py`,
                  a .env / beartype / jaxtyping / cuEquivariance bootstrap, and
                  `foundry.utils.ddp`, which is Lightning-based, are replaced by
                  minimal stubs, so no Lightning, hydra or environs dependency is
                  pulled in.

9. PXDesign — binder design on a Protenix diffusion module with the trunk removed.
   An independent ttnn reimplementation under tt_bio/pxdesign/; no upstream compute
   code is vendored.
   Source:        github.com/bytedance/PXDesign
   License:       Apache-2.0 — Copyright (c) ByteDance Ltd.
   Weights:       pxdesign_v0.1.0.pt, downloaded on first use from
                  pxdesign.tos-cn-beijing.volces.com (Apache-2.0, not redistributed here)

10. AlphaFold2 monomer pTM — the reference model and parameters behind the AF2-IG
   selection stage that scores PXDesign's designs. tt_bio/af2_reference.py is a torch
   transcription of the AlphaFold model code that ColabDesign 1.1.3 vendors
   (colabdesign/af/alphafold/model/: modules, common_modules, all_atom, r3), so it is a
   modified port of DeepMind's code, not an independent design. The featurizer and the
   rest of tt_bio/af2*.py are tt-bio's own.
   Source:        github.com/google-deepmind/alphafold via github.com/sokrypton/ColabDesign
   License:       Apache-2.0 — Copyright 2021 DeepMind Technologies Limited (code);
                  CC BY 4.0 — Copyright (c) DeepMind Technologies Limited (parameters,
                  storage.googleapis.com/alphafold/alphafold_params_2022-12-06.tar)
   Modifications: translated from JAX/Haiku to torch, inference only, reading the
                  checkpoint through tt_bio.af2_weights.
   Note:          only params_model_1_ptm.npz is extracted and kept; downloaded on first
                  use, not redistributed here

Not included

11. BindCraft 2 — tt-bio contains none of its code and does not distribute or serve it.
   tt_bio.bindcraft2 is an adapter that binds to a copy of BindCraft 2 you install
   yourself, and runs that copy's own design loop with tt-bio's AlphaFold 2 trunk.
   BindCraft 2 stays under its own licence; tt-bio grants no rights in it.
   Source:        github.com/PacesaLab/BindCraft2
   License:       BindCraft2 Source-Available License (Hosting-Restricted) —
                  Copyright (c) 2026 Martin Pacesa, University of Zurich, and the
                  BindCraft2 contributors. Not an open source licence. Read at
                  commit a8d0f2002df3 (licence text last changed ee849bcd52d5).
   Restriction:   offering BindCraft 2, a modified version, or a service whose
                  material functionality comes from it to third parties as a
                  hosted, API, web, workflow-platform or agent-tool service needs a
                  separate written commercial licence from the copyright holders.
                  That includes running it through tt-bio on a Galaxy for others.
                  Running it yourself and sharing the designs it produces does not.
   Name:          "BindCraft2" is a trademark of the University of Zurich. tt-bio is
                  not affiliated with or endorsed by the BindCraft 2 authors.
