generated: '2026-08-22'
method: searched
source: https://github.com/HorizonRobotics/OE-Skills — horizon/docs/oe_code_chunk_*.md, plus public registry probes
summary: >-
  Horizon Robotics ships a substantial first-party Python/C++ SDK surface, but NONE of it
  is published to a public package registry. Every library below is distributed only
  inside the versioned OpenExplorer (OE) release package — a tarball or Docker image
  downloaded from the Horizon developer portal behind an account. npm, PyPI, Maven
  Central, NuGet, pkg.go.dev, RubyGems, Packagist and crates.io were all probed and
  returned 404 for every candidate name, so there is no registry metadata endpoint to
  read a version or a publication date from. `version: null` on each entry means
  "checked, nothing published", not "not checked". A consumer cannot pin a Horizon SDK
  by version from any public index, and cannot tell from outside whether the copy they
  hold is current.
registry_probe:
  date: '2026-08-22'
  registries_probed: [npm, pypi, maven-central, nuget, pkg.go.dev, rubygems, packagist, crates.io]
  names_probed:
  - horizon-plugin-pytorch
  - horizon_plugin_pytorch
  - hbdk
  - hbdk4
  - hbdk4-compiler
  - hmct
  - horizon-tc-ui
  - horizon-nn
  - hobot-dnn
  - horizon-robotics
  hits: 0
packages:
- language: python
  registry: vendor
  name: horizon_plugin_pytorch
  official: true
  version: null
  published: null
  url: https://doc.oe.horizon.auto/
  install: bundled in the OpenExplorer (OE) release package
  summary: >-
    Horizon's PyTorch quantization plugin — QAT preparation, fake-quantize configuration,
    march selection, quantize/dequantize insertion, export to HBDK.
  note: Not on PyPI (404 for both hyphen and underscore spellings on 2026-08-22).
- language: python
  registry: vendor
  name: horizon_plugin_profiler
  official: true
  version: null
  published: null
  url: https://doc.oe.horizon.auto/
  install: bundled in the OpenExplorer (OE) release package
  summary: Profiling and training/deployment consistency debugging companion to horizon_plugin_pytorch.
  note: Not on PyPI.
- language: python
  registry: vendor
  name: hbdk4-compiler
  official: true
  version: null
  published: null
  url: https://doc.oe.horizon.auto/
  install: bundled in the OpenExplorer (OE) release package
  summary: HBDK4 compiler API — model convert/compile/export/serialize, HBM inspection and modification, node insertion/removal.
  note: Not on PyPI.
- language: python
  registry: vendor
  name: hmct
  official: true
  version: null
  published: null
  url: https://doc.oe.horizon.auto/
  install: bundled in the OpenExplorer (OE) release package
  summary: Horizon Model Conversion Toolkit — PTQ calibration, mixed-precision and cosine-similarity accuracy tuning. A CUDA variant (hmct-gpu) is documented alongside it.
  note: Not on PyPI.
- language: python
  registry: vendor
  name: horizon_tc_ui
  official: true
  version: null
  published: null
  url: https://doc.oe.horizon.auto/
  install: bundled in the OpenExplorer (OE) release package
  summary: The toolchain CLI package (hb_compile, hb_model_info, hb_verifier, hb_analyzer, hb_config_generator, hb_eval_preprocess). See cli/horizon-robotics-cli.yml.
  note: Not on PyPI.
- language: c++
  registry: vendor
  name: hbDNN
  official: true
  version: null
  published: null
  url: https://doc.oe.horizon.auto/
  install: bundled in the OpenExplorer (OE) / UCP on-board runtime
  summary: The on-board BPU inference runtime API used to load and execute compiled .hbm models on Journey silicon.
  note: C++ headers + shared libraries shipped with the board BSP; no public artifact repository.
- language: python
  registry: vendor
  name: llm_compression
  official: true
  version: null
  published: null
  url: https://github.com/HorizonRobotics/OE-Skills/blob/main/horizon/docs/oe_code_chunk_llm_compression.md
  install: bundled in the OE LLM package
  summary: LLM calibration, compression and on-board evaluation pipeline (LightCompress RTN/GPTQ/AWQ/SmoothQuant).
  note: Not on PyPI; installed by the OE-Skills `oe-llm-package-install` skill.
- language: c++
  registry: vendor
  name: oellm_runtime
  official: true
  version: null
  published: null
  url: https://github.com/HorizonRobotics/OE-Skills/blob/main/horizon/docs/oe_code_chunk_oellm_runtime.md
  install: bundled in the OE LLM package
  summary: On-board runtime for compressed LLMs on Journey/BPU targets.
  note: Not on PyPI.
open_source_repositories:
  note: >-
    github.com/HorizonRobotics publishes 44 research and robotics repositories (ALF,
    Sparse4D, HoloMotion, EmbodiedGen, RoboOrchard, BIP3D, GUMP, ...). These are research
    code released under open licences, not versioned client SDKs for a Horizon API, so
    they are catalogued as source rather than as packages.
  url: https://github.com/HorizonRobotics
x-evidence:
  fetched: '2026-08-22'
  urls:
  - url: https://pypi.org/pypi/horizon-plugin-pytorch/json
    status: 404
  - url: https://pypi.org/pypi/hbdk4-compiler/json
    status: 404
  - url: https://pypi.org/pypi/hmct/json
    status: 404
  - url: https://pypi.org/pypi/horizon-tc-ui/json
    status: 404
  - url: https://registry.npmjs.org/horizon-robotics
    status: 404
  - url: https://api.github.com/repos/HorizonRobotics/OE-Skills/git/trees/main?recursive=1
    status: 200