Skip to content

armgit / hypit

Public Imported from GitHub
0

Clone any viral video with AI agents. Not just a script, the whole workflow: swap the face, the words, the B-roll, ship 100 variants in one command, and get your 100M views.

docs(whisperx): explain explicit proxy trust during preparation

kashorinde2a5619a51c
diff --git a/packages/provider-whisperx-local/README.md b/packages/provider-whisperx-local/README.md
index 4300c846..ffa8243a 100644
--- a/packages/provider-whisperx-local/README.md
+++ b/packages/provider-whisperx-local/README.md
@@ -96,6 +96,9 @@ network and cache variables there before `programs up` or `runtime up`. A servic
retains its earlier environment. Inspect its reported log before deciding whether a selected
Program needs restarting, and account for active work using it.
+If NLTK refuses a proxied fetch during preparation, follow the service’s
+[explicit proxy preparation](../../services/whisperx/README.md#preparing-sentence-data-through-a-proxy).
+
Preparation commands write `install.log`; the running service writes `program.log`, with stderr in
`program.err.log` on Windows. Inspect the stderr file for Python model-loading and download messages.
`programs status` reports these files as `installationLogPath`, `logPath` and `errorLogPath` when they
diff --git a/services/whisperx/README.md b/services/whisperx/README.md
index 76743cbb..7aa2cdc8 100644
--- a/services/whisperx/README.md
+++ b/services/whisperx/README.md
@@ -91,6 +91,26 @@ Transcripts and audio content are not included in these service progress entries
after ASR loading; it does not report all language caches as ready. A first request can load a
prepared aligner into memory, but cannot download it.
+## Preparing sentence data through a proxy
+
+NLTK's downloader refuses a proxied request unless the operator explicitly trusts that proxy:
+it cannot enforce its direct-connection address checks through a proxy. This can stop preparation
+while fetching the NLTK data index, even when uv and Hugging Face downloads work.
+
+For a proxy you trust, scope NLTK's native opt-in to the explicit preparation command:
+
+```bash
+NLTK_ALLOW_PROXIED_URLOPEN=1 hypit programs prepare --endpoint whisperx.local
+```
+
+Use your selected Endpoint instance id. On PowerShell, set `$env:NLTK_ALLOW_PROXIED_URLOPEN = "1"`
+for that preparation session, then restore its previous value. The preparation process already
+inherits its caller's environment; no Hypit-specific proxy flag is needed. This setting authorizes
+NLTK to use the configured proxy; it neither selects a mirror nor enables inference downloads.
+Without that trust choice, use a deliberately selected direct route or supply the sentence data
+in the selected NLTK data directory before preparation. Hypit does not change the choice automatically.
+Existing usable sentence data is read locally without refreshing the download index.
+
## Package preparation
Local package preparation includes `src/**/*.py` in uv's package cache inputs. Updating service code
diff --git a/skills/hypit/references/environment/local-tools.md b/skills/hypit/references/environment/local-tools.md
index de6e9a4a..4200212a 100644
--- a/skills/hypit/references/environment/local-tools.md
+++ b/skills/hypit/references/environment/local-tools.md
@@ -225,6 +225,7 @@ Mirrors address particular download clients and hosts:
| Python packages | pip uses `--index-url` / `PIP_INDEX_URL`; uv uses `--default-index` / `UV_DEFAULT_INDEX`. They are different clients. Hypit's managed service uses frozen uv dependencies; consult its Provider README before expecting an index change to redirect locked artifact URLs. |
| Python runtime | uv's `UV_PYTHON_INSTALL_MIRROR` selects a compatible Python-distribution mirror. A PyPI mirror does not supply Python itself. An already compatible installed Python may avoid this download. |
| Hugging Face weights | `HF_ENDPOINT` selects a compatible Hub endpoint; `HF_HOME` / `HF_HUB_CACHE` select reusable cache locations. A model's redirected weight host and its language-alignment download must also be reachable. |
+| NLTK sentence data | If preparation reports `NLTK_ALLOW_PROXIED_URLOPEN`, the downloader needs an explicit trust decision for the configured proxy. For a trusted proxy, set that native variable to `1` only on the preparation command; do not switch it on automatically or change inference. |
| HyperFrames browser | The selected Provider owns the archive source; follow [browser preparation](#prepare-the-local-rendering-browser) and its installed README. |
| FFmpeg and other binaries | Use the selected package manager's binary-download settings or a compatible official prebuilt installation. An npm/PyPI mirror does not generally redirect these downloads. Homebrew bottles and GitHub release assets have their own sources. |