boson-video 0.2.0__tar.gz

boson_video-0.2.0/.dockerignore

@@ -0,0 +1,11 @@
+.env
+.git
+.venv
+.pytest_cache
+.claude
+__pycache__
+*.pyc
+out
+tests
+docs
+scripts

boson_video-0.2.0/.gitignore

@@ -0,0 +1,8 @@
+.venv/
+__pycache__/
+*.pyc
+.pytest_cache/
+out/
+.env
+.DS_Store
+.claude/

boson_video-0.2.0/AGENTS.md

@@ -0,0 +1,76 @@
+# Boson-Video: guide for coding agents
+
+Read any video like a document. **Read [docs/DIRECTION.md](docs/DIRECTION.md) first.** It
+holds the agreed direction, the milestones, who does what, and the owner's open
+decisions. The target screen is [docs/read-view.html](docs/read-view.html); open it in a
+browser.
+
+## Commands
+
+- `uv sync` once, then `uv run boson-video <youtube link | id | file> [--open] [--no-words] [--no-summary] [--lang zh_CN] [-o folder]` (videos are kept in `BOSON_VIDEO_HOME`, default `~/.boson-video`, for the CLI, the page and the plugin alike)
+- `uv run boson-video ask <video> "question"`: the moment that answers it (needs the page to exist)
+- `uv run boson-video serve [--open [<video>]]`: the website on 127.0.0.1:8765: paste a link, read it with the video playing beside it and a working ask box
+- `uv run boson-video web [--hosted] [--host 0.0.0.0] [--port 8770]`: the product page, invite-only, new design (the first run prints your code); `--hosted` refuses YouTube links
+- `uv run boson-video models [zh|en|all]`: fetch the local speech models now (otherwise the first video that needs one downloads it)
+- `uv run boson-video invite [label] [--per-day 10] [--asks 100] | --list | --revoke CODE`: invite codes for it
+- `uv run boson-video mcp`: the plugin (MCP over stdio) for Claude Code, Claude Desktop, Cursor, Codex; setup in `docs/plugin.md`
+- `uv run python scripts/mcp_smoke.py <video> [--no-keys]`: run the plugin like an AI app and call every tool
+- `uv run pytest` (offline) · `uv run pytest -m live` (hits YouTube, runs Apple's transcriber)
+- `uv run python scripts/accuracy.py <video id> <caption language> <locale>`: score our transcript against human captions
+
+## Map
+
+- `src/boson_video/youtube.py`: watch page → video, storyboard levels, chapters, heatmap, caption list
+- `storyboard.py`: fetch sheets in parallel, slice into frames
+- `local.py`: ffmpeg keyframes → frames + our own sheets
+- `audio.py`: yt-dlp audio download, 16 kHz WAV, pauses, cutting into pieces
+- `speech.py` + `bv_speech.swift`: Apple's on-device transcriber (macOS 26), built on first use into
+  `~/Library/Caches/boson-video/`; pieces run at once; `names()` lists the video's own names for the writer
+- `sensevoice.py`: SenseVoice through sherpa-onnx, where Apple's transcriber isn't available (Windows, Linux)
+- `mercury.py`: the one Mercury client (strict JSON, retries, cost)
+- `study.py`: English for every passage, the glossary, and answers to questions (Jev finds, Mercury explains, Jev checks)
+- `web.py`: `boson-video web`, the product page (Starlette): invite codes and a signed cookie, per-code videos and daily limits, links and uploads, the document as JSON, answers with frames · `app/`: its one page (`index.html`, `app.css`, `app.js`; plain JS, no libraries, no product name): ribbon with hover preview and readout; Summary, Transcript, Terms, Scenes and Ask tabs; original / both / English
+- `server.py`: `boson-video serve`, the website: a front page to paste a YouTube link (`/api/open` starts the background build, `/api/status` follows it), pages and their frames, `/api/ask` (JSON from its own origin only), questions kept in `<id>/notes.json`
+- `library.py`: where videos live (`BOSON_VIDEO_HOME`), load and save (page, `timeline.json`, `sheets/`)
+- `jobs.py`: builds a video in a background thread; progress in `status.json`
+- `frames.py`: picks moments from the scene map; full-resolution frames (yt-dlp stream address + ffmpeg, one per second needed); thumbnail fallback
+- `screens.py`: what was shown, as text: frames at the scene map's moments, read by OCR; subtitles kept apart; build steps credited with their new lines
+- `ocr.py`: RapidOCR in a worker process (never alongside sherpa-onnx), reading frames as they arrive
+- `moments.py`: the moment, the unit everything cites: scenes, words, screen text and the frame joined, coded state, delta, trajectory or seek
+- `plugin.py`: the plugin's tools as plain functions (briefing, read, frames, search, check, list) · `mcp_server.py`: wires them to MCP
+- `accuracy.py`: error rates against human captions (words; characters for Chinese)
+- `writer.py`: Mercury writes the summary and sections (strict JSON schema, every sentence cites passages, one retry)
+- `checker.py`: Jev checks each sentence against its passages; numbers compared in code; citations repaired
+- `ask.py`: Jev finds the passage that answers a question (one pass, or two for more than 255 passages)
+- `env.py`: loads keys from `.env`
+- `scenes.py`: cuts, looks, base/repeat/new, `profile()` headline
+- `render.py`: the self-contained read view (ribbon, summary with language switch, search, scenes); a little JS, no libraries
+- `pipeline.py`: `build` (scenes), `add_words` (speech), `add_summary` (write + check), with per-stage timings · `cli.py`: entry point
+- `timeline.py`: the shared data model; `timeline.json` is the contract between stages and for other tools, versioned, spec in `docs/timeline-format.md`
+- `docs/`: `DIRECTION.md` (source of truth) and `read-view.html` (target design); `deploy.md`: how the owner hosts it (Dockerfile, Coolify, the Vercel address), kept out of the README
+
+## Rules
+
+- Follow `docs/DIRECTION.md`: build the next milestone, respect "Not now", and update the
+  milestone table when work lands. Don't change the direction without the owner.
+- Speed is the product. Time every stage (`Stopwatch` in `pipeline.py`) and update the
+  measured tables in `docs/measured.md` when numbers move. Say which numbers are measured and which
+  are estimates.
+- Scene thresholds were calibrated on three reference videos. After any change to
+  `scenes.py`, recheck all three: 9JKT5rBbrwM (talking head, expect 3 scenes),
+  zjkBMFhNj_g (slides + webcam box, about 59 new, no false repeats), dQw4w9WgXcQ (fast
+  cuts). `tests/test_scenes.py` encodes the same cases.
+- Jev is text-only and doesn't write text: it judges (checks, finds) and another model
+  writes. It leans toward the option listed first, so every Choice goes out twice in the same
+  request with the options reversed, and the answers are averaged (`checker.both_orders`).
+- The engine must work with no keys. Anything that needs Jev or a writing model is an extra
+  that says what it is and what checked it. Read TypeSafe's live docs (https://docs.typesafe.ai/llms.txt) before
+  integrating it.
+- Keys live in `.env`, which git ignores: `TYPESAFE_API_KEY` (Jev), `INCEPTION_API_KEY`
+  (Mercury). Never commit keys.
+- Don't try to get around YouTube's bot checks (PO tokens, player API clients). The
+  Chrome extension is the route for that.
+- Ask the owner before downloading models or media onto the owner's machines. `docs/DIRECTION.md` lists what is pending.
+  (On a user's computer the plugin downloads its speech model on first use and says so: that is the product.)
+- Publishing: `uv build`, then `uv publish` (PyPI) and `mcp-publisher publish` (the registry, from `server.json`).
+  Keep the version the same in `pyproject.toml`, `src/boson_video/__init__.py` and `server.json`.

boson_video-0.2.0/CLAUDE.md

@@ -0,0 +1,14 @@
+# Boson-Video
+
+Read any video like a document. The agreed direction and the agent guide are imported
+below; read them before building.
+
+@docs/DIRECTION.md
+@AGENTS.md
+
+## Claude Code notes
+
+- For Jev / TypeSafe work, use the `typesafe-ai` skill.
+- The target design is also published as an artifact (owner-only):
+  https://claude.ai/artifact/4KTYhto8fKhLPptrokXyUn. After editing `docs/read-view.html`,
+  republish it with that URL so the link stays the same.

boson_video-0.2.0/Dockerfile

@@ -0,0 +1,24 @@
+# boson-video web on a server (Coolify, or any Docker host). It runs with --hosted: uploads only,
+# because YouTube downloads stay on the user's side (docs/DIRECTION.md). Mount a persistent
+# volume at /data: videos, invite codes and the speech models (about 650 MB, fetched on the
+# first start by deploy/start.sh) live there. Keys come from the environment, never the image.
+FROM python:3.12-slim
+
+# ffmpeg reads the uploads; libgl1 and libglib2.0-0 are what OpenCV (under the screen reader) needs
+RUN apt-get update \
+ && apt-get install -y --no-install-recommends ffmpeg ca-certificates curl bzip2 libgl1 libglib2.0-0 \
+ && rm -rf /var/lib/apt/lists/*
+COPY --from=ghcr.io/astral-sh/uv:latest /uv /usr/local/bin/uv
+
+WORKDIR /app
+COPY pyproject.toml uv.lock README.md ./
+COPY src ./src
+RUN uv sync --frozen --no-dev
+COPY deploy/start.sh ./start.sh
+
+ENV BOSON_VIDEO_HOME=/data \
+    BOSON_MODELS=/data/models \
+    PATH="/app/.venv/bin:$PATH" \
+    PYTHONUNBUFFERED=1
+EXPOSE 8770
+CMD ["sh", "/app/start.sh"]

boson_video-0.2.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 James Lin
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

boson_video-0.2.0/PKG-INFO

@@ -0,0 +1,130 @@
+Metadata-Version: 2.5
+Name: boson-video
+Version: 0.2.0
+Summary: Skim any video: a timed summary, a bilingual transcript, screen text and checked answers, as a web page or an MCP plugin.
+Project-URL: Homepage, https://github.com/linboxin/Boson-Video
+Project-URL: Repository, https://github.com/linboxin/Boson-Video
+Project-URL: Issues, https://github.com/linboxin/Boson-Video/issues
+Author: James Lin
+License-Expression: MIT
+License-File: LICENSE
+Keywords: claude,learning,mcp,mcp-server,ocr,subtitles,summary,transcript,video,youtube
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: End Users/Desktop
+Classifier: Natural Language :: Chinese (Simplified)
+Classifier: Natural Language :: English
+Classifier: Operating System :: MacOS
+Classifier: Operating System :: Microsoft :: Windows
+Classifier: Operating System :: POSIX :: Linux
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Multimedia :: Video
+Classifier: Topic :: Text Processing
+Requires-Python: >=3.12
+Requires-Dist: deno>=2.0
+Requires-Dist: httpx[http2]>=0.28
+Requires-Dist: imageio-ffmpeg>=0.6.0
+Requires-Dist: mcp>=2.3.0
+Requires-Dist: numpy>=2.0
+Requires-Dist: pillow>=11.0
+Requires-Dist: rapidocr-onnxruntime>=1.4.4
+Requires-Dist: sherpa-onnx-core>=1.13.8
+Requires-Dist: sherpa-onnx>=1.13.8
+Requires-Dist: typesafe-sdk>=0.7.2
+Requires-Dist: yt-dlp[default]>=2026.8.19
+Description-Content-Type: text/markdown
+
+# Boson-Video
+
+**Get the point of any video without watching all of it.** Paste a YouTube link or drop a video
+file. Every line of the summary, the transcript and the screen text carries the second it came
+from, and questions are answered from the video and checked against it.
+
+<!-- mcp-name: io.github.linboxin/boson-video -->
+
+![A video read in Boson-Video: the player and the ribbon on the left, the terms explained on the right](https://raw.githubusercontent.com/linboxin/Boson-Video/main/docs/screenshot.jpg)
+
+- **The ribbon:** the whole video on one strip (scenes, chapters, most replayed). Hover to see any second's frame and words.
+- **Summary:** sections with their frames; every sentence timed and checked (✓ ? ✗).
+- **Transcript:** the original with English beneath, technical terms explained, on-screen text read in.
+- **Ask:** answers that cite their moments and show the frames, with background kept apart.
+- **Your own AI:** the same document as an MCP plugin for Claude, Cursor, Codex and more.
+
+Built for long talks, lectures and finance videos in a language you half know (Chinese and English today).
+
+## In your own AI
+
+One line, nothing else to install ([uv](https://docs.astral.sh/uv/) runs it):
+
+```bash
+claude mcp add boson-video -- uvx boson-video mcp
+```
+
+Claude Desktop (`claude_desktop_config.json`) or Cursor (`.cursor/mcp.json`):
+
+```json
+{ "mcpServers": { "boson-video": { "command": "uvx", "args": ["boson-video", "mcp"] } } }
+```
+
+Codex (`~/.codex/config.toml`):
+
+```toml
+[mcp_servers.boson-video]
+command = "uvx"
+args = ["boson-video", "mcp"]
+```
+
+Then ask your AI about any YouTube link. It opens the video, reads the transcript, looks at the
+frames that matter at full resolution, searches, and checks its claims, citing every moment. No
+keys needed. Everything runs on your computer. Details: [docs/plugin.md](https://github.com/linboxin/Boson-Video/blob/main/docs/plugin.md).
+
+## Run the page
+
+```bash
+uvx boson-video web             # http://127.0.0.1:8770; the first run prints your invite code
+```
+
+Keys make it fuller: `INCEPTION_API_KEY` (Mercury writes the summary, the English and the terms)
+and `TYPESAFE_API_KEY` (Jev checks every sentence and answers questions). Without them you still get
+the scenes, the transcript and search.
+
+## Tested videos
+
+Measured, one run each unless a range is shown. Mac: M5 MacBook with Apple's transcriber. Windows: a 12-thread laptop with
+SenseVoice or Parakeet on the CPU. Full tables: [docs/measured.md](https://github.com/linboxin/Boson-Video/blob/main/docs/measured.md).
+
+| Video | Length | Scene map | Words | Transcript errors¹ | Screen text | Summary ✓ |
+| --- | --- | --- | --- | --- | --- | --- |
+| Money or Life 美股频道, Meta and AI (zh, talking head) | 24:53 | 1.3 s | 13.8 s (Mac) | no human captions | — | 13 / 13 |
+| Money or Life 美股频道, AI drug discovery (zh, slides) | 28:24 | 1.1 s | 18.5 s (Mac) | not scored | 36 moments read | 15 / 16 |
+| 程序员老王, LLM abliteration (zh, animated diagrams) | 11:35 | — | 10.0 s (Windows) | not scored | 88 moments; subtitles apart at 85 | 21–22 / 22–23 (several runs) |
+| 陳永儀, TEDxTaipei (zh, talk) | 14:29 | — | 7.1 s (Mac) | 5.3% chars (Mac), 4.8% (SenseVoice) | — | 14 / 14 |
+| Ken Robinson, TED (en, talk) | 20:06 | — | 9.9 s (Mac) | 10.0% words (Mac), 10.5% (Parakeet) | — | 22 / 25 |
+| Sean's AI Stories, agent observability (en, screen recording) | 20:48 | 1.5 s | 13.2 s (Mac) | not scored | none: YouTube refused full resolution | 23 / 26 |
+| Andrej Karpathy, Intro to LLMs (en, slides) | 59:48 | 1.6 s | 30.5 s (Mac) | only automatic captions | 19 of 20 slides right | 22 / 23 |
+| Rick Astley, Never Gonna Give You Up (fast cuts) | 3:33 | 0.7 s | — | — | — | — |
+
+¹ Against human captions (`scripts/accuracy.py`); about half the English errors are filler words the captions leave out.
+Summary ✓: sentences confirmed against what was said (Jev and code).
+
+## Command line
+
+```bash
+uvx boson-video <link | id | file>        # build a video: scenes, words, screen text, summary
+uvx boson-video ask <video> "question"    # the moment that answers it
+uvx boson-video models                    # fetch the local speech models now instead of on first use
+```
+
+From a clone: `uv sync`, then `uv run boson-video …`; `uv run pytest` runs the tests (offline).
+
+## More
+
+- [docs/workflow.md](https://github.com/linboxin/Boson-Video/blob/main/docs/workflow.md): how the workflow is designed: stages, the document, the ways in
+- [docs/how-it-works.md](https://github.com/linboxin/Boson-Video/blob/main/docs/how-it-works.md): the pipeline, what YouTube allows, limits
+- [docs/DIRECTION.md](https://github.com/linboxin/Boson-Video/blob/main/docs/DIRECTION.md): what we're building, milestones, decisions
+- [docs/timeline-format.md](https://github.com/linboxin/Boson-Video/blob/main/docs/timeline-format.md): the document format, for other tools
+- [AGENTS.md](https://github.com/linboxin/Boson-Video/blob/main/AGENTS.md): the guide for coding agents
+
+MIT licensed. YouTube's terms don't allow automated access for products, so everything that touches
+YouTube runs on your own computer.

boson_video-0.2.0/README.md

@@ -0,0 +1,93 @@
+# Boson-Video
+
+**Get the point of any video without watching all of it.** Paste a YouTube link or drop a video
+file. Every line of the summary, the transcript and the screen text carries the second it came
+from, and questions are answered from the video and checked against it.
+
+<!-- mcp-name: io.github.linboxin/boson-video -->
+
+![A video read in Boson-Video: the player and the ribbon on the left, the terms explained on the right](https://raw.githubusercontent.com/linboxin/Boson-Video/main/docs/screenshot.jpg)
+
+- **The ribbon:** the whole video on one strip (scenes, chapters, most replayed). Hover to see any second's frame and words.
+- **Summary:** sections with their frames; every sentence timed and checked (✓ ? ✗).
+- **Transcript:** the original with English beneath, technical terms explained, on-screen text read in.
+- **Ask:** answers that cite their moments and show the frames, with background kept apart.
+- **Your own AI:** the same document as an MCP plugin for Claude, Cursor, Codex and more.
+
+Built for long talks, lectures and finance videos in a language you half know (Chinese and English today).
+
+## In your own AI
+
+One line, nothing else to install ([uv](https://docs.astral.sh/uv/) runs it):
+
+```bash
+claude mcp add boson-video -- uvx boson-video mcp
+```
+
+Claude Desktop (`claude_desktop_config.json`) or Cursor (`.cursor/mcp.json`):
+
+```json
+{ "mcpServers": { "boson-video": { "command": "uvx", "args": ["boson-video", "mcp"] } } }
+```
+
+Codex (`~/.codex/config.toml`):
+
+```toml
+[mcp_servers.boson-video]
+command = "uvx"
+args = ["boson-video", "mcp"]
+```
+
+Then ask your AI about any YouTube link. It opens the video, reads the transcript, looks at the
+frames that matter at full resolution, searches, and checks its claims, citing every moment. No
+keys needed. Everything runs on your computer. Details: [docs/plugin.md](https://github.com/linboxin/Boson-Video/blob/main/docs/plugin.md).
+
+## Run the page
+
+```bash
+uvx boson-video web             # http://127.0.0.1:8770; the first run prints your invite code
+```
+
+Keys make it fuller: `INCEPTION_API_KEY` (Mercury writes the summary, the English and the terms)
+and `TYPESAFE_API_KEY` (Jev checks every sentence and answers questions). Without them you still get
+the scenes, the transcript and search.
+
+## Tested videos
+
+Measured, one run each unless a range is shown. Mac: M5 MacBook with Apple's transcriber. Windows: a 12-thread laptop with
+SenseVoice or Parakeet on the CPU. Full tables: [docs/measured.md](https://github.com/linboxin/Boson-Video/blob/main/docs/measured.md).
+
+| Video | Length | Scene map | Words | Transcript errors¹ | Screen text | Summary ✓ |
+| --- | --- | --- | --- | --- | --- | --- |
+| Money or Life 美股频道, Meta and AI (zh, talking head) | 24:53 | 1.3 s | 13.8 s (Mac) | no human captions | — | 13 / 13 |
+| Money or Life 美股频道, AI drug discovery (zh, slides) | 28:24 | 1.1 s | 18.5 s (Mac) | not scored | 36 moments read | 15 / 16 |
+| 程序员老王, LLM abliteration (zh, animated diagrams) | 11:35 | — | 10.0 s (Windows) | not scored | 88 moments; subtitles apart at 85 | 21–22 / 22–23 (several runs) |
+| 陳永儀, TEDxTaipei (zh, talk) | 14:29 | — | 7.1 s (Mac) | 5.3% chars (Mac), 4.8% (SenseVoice) | — | 14 / 14 |
+| Ken Robinson, TED (en, talk) | 20:06 | — | 9.9 s (Mac) | 10.0% words (Mac), 10.5% (Parakeet) | — | 22 / 25 |
+| Sean's AI Stories, agent observability (en, screen recording) | 20:48 | 1.5 s | 13.2 s (Mac) | not scored | none: YouTube refused full resolution | 23 / 26 |
+| Andrej Karpathy, Intro to LLMs (en, slides) | 59:48 | 1.6 s | 30.5 s (Mac) | only automatic captions | 19 of 20 slides right | 22 / 23 |
+| Rick Astley, Never Gonna Give You Up (fast cuts) | 3:33 | 0.7 s | — | — | — | — |
+
+¹ Against human captions (`scripts/accuracy.py`); about half the English errors are filler words the captions leave out.
+Summary ✓: sentences confirmed against what was said (Jev and code).
+
+## Command line
+
+```bash
+uvx boson-video <link | id | file>        # build a video: scenes, words, screen text, summary
+uvx boson-video ask <video> "question"    # the moment that answers it
+uvx boson-video models                    # fetch the local speech models now instead of on first use
+```
+
+From a clone: `uv sync`, then `uv run boson-video …`; `uv run pytest` runs the tests (offline).
+
+## More
+
+- [docs/workflow.md](https://github.com/linboxin/Boson-Video/blob/main/docs/workflow.md): how the workflow is designed: stages, the document, the ways in
+- [docs/how-it-works.md](https://github.com/linboxin/Boson-Video/blob/main/docs/how-it-works.md): the pipeline, what YouTube allows, limits
+- [docs/DIRECTION.md](https://github.com/linboxin/Boson-Video/blob/main/docs/DIRECTION.md): what we're building, milestones, decisions
+- [docs/timeline-format.md](https://github.com/linboxin/Boson-Video/blob/main/docs/timeline-format.md): the document format, for other tools
+- [AGENTS.md](https://github.com/linboxin/Boson-Video/blob/main/AGENTS.md): the guide for coding agents
+
+MIT licensed. YouTube's terms don't allow automated access for products, so everything that touches
+YouTube runs on your own computer.

boson_video-0.2.0/deploy/start.sh

@@ -0,0 +1,21 @@
+#!/bin/sh
+# Start boson-video web in a container: fetch the speech models into the volume once (SenseVoice
+# for Chinese, Parakeet for English, Silero for finding speech; about 650 MB), then serve.
+set -e
+M="${BOSON_MODELS:-/data/models}"
+URL=https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models
+mkdir -p "$M"
+
+fetch() {
+  [ -d "$M/$1" ] && return 0
+  echo "downloading $1 (once, into the volume)"
+  rm -rf "$M/.part" && mkdir -p "$M/.part"
+  curl -fsSL "$URL/$1.tar.bz2" | tar -xj -C "$M/.part"
+  mv "$M/.part/$1" "$M/$1" && rm -rf "$M/.part"
+}
+fetch sherpa-onnx-sense-voice-zh-en-ja-ko-yue-int8-2025-09-09
+fetch sherpa-onnx-nemo-parakeet-tdt-0.6b-v2-int8
+[ -f "$M/silero_vad.onnx" ] || curl -fsSL -o "$M/silero_vad.onnx" "$URL/silero_vad.onnx"
+
+# Behind Coolify's proxy, which brings HTTPS and the domain; the first start prints your invite code.
+exec boson-video web --hosted --host 0.0.0.0 --port "${PORT:-8770}"

boson_video-0.2.0/deploy/vercel/vercel.json

@@ -0,0 +1,7 @@
+{
+  "$schema": "https://openapi.vercel.sh/vercel.json",
+  "rewrites": [
+    { "source": "/", "destination": "https://h0wsk84kc00084ccg08o040k.40.160.231.143.sslip.io/" },
+    { "source": "/:path*", "destination": "https://h0wsk84kc00084ccg08o040k.40.160.231.143.sslip.io/:path*" }
+  ]
+}