> For the complete documentation index, see [llms.txt](https://getsquish.gitbook.io/squish/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://getsquish.gitbook.io/squish/resources/faq.md). # FAQ ## Does Squish use audio? The CLI and local MCP server (0.3+) extract a low-rate audio-activity envelope and align it to the same absolute timeline as the frames. It can reveal *when* energy changes — useful when the picture barely changes — but it does not play, transcribe, classify, or interpret audio. Levels use one `clip_peak` scale for the full source, including zoomed runs, so quiet and loud windows remain comparable. Videos with no audio track still work. The web app, hosted API, and remote MCP remain visual-only until separately released. For dialogue or "what made that sound?", you still need a transcript or audio-capable model. Those tools compose well. Squish timecodes are absolute to the source video, and transcripts are timestamped on that same time axis — so a contact sheet plus a transcript line up cell-for-line, and a model given both can connect what was said to what was visible. ## Why not just upload the video to the model? Sometimes you should. But many models and plans are image-only, video uploads are slow and token-expensive, and you can't inspect what the model actually "saw". A contact sheet is compact, works with **any** vision model (including local ones in LM Studio or Ollama), is inspectable — you see exactly the frames the model sees — and pastes anywhere: chat, docs, bug reports. See [video as an address space](/squish/the-primitive/video-as-address-space.md) for the deeper reason: most questions touch a tiny fraction of the timeline, so navigating beats replaying. ## What does it cost? Everything in this repo is free: the CLI and MCP server run entirely on your machine, at every density, with no account and no key — that's permanent, not a trial. The only paid path is the optional [hosted API](/squish/getting-started/quickstart-api.md) at `api.getsquish.app`, which runs on prepaid credits for the cases where local processing isn't an option (CI, serverless, no ffmpeg); accounts that have never purchased get a small free daily allowance, applied automatically on the first request. ## Is Squish open source? The engine is — Apache-2.0, published as [`@getsquish/squish`](https://www.npmjs.com/package/@getsquish/squish) on npm with the source at [github.com/getsquish/squish](https://github.com/getsquish/squish). That repo is a curated mirror of a private monorepo (the web app and hosted API live there); issues and PRs are welcome on the public repo — see [Contributing](/squish/resources/contributing.md). The Squish name, logo, mascot, and brand assets are reserved and not part of the license. ## What video formats work? Whatever your ffmpeg decodes. The CLI and MCP server shell out to your system `ffmpeg`/`ffprobe`, so format support is exactly your local build's — in practice that covers MP4, MOV, WebM, MKV, and most everything else. If `ffprobe` can read a duration from the file, Squish can sheet it. ## How accurate are the timecodes? Frame-accurate on modern ffmpeg. The engine uses input seeking (`-ss` before `-i`): it jumps to the nearest keyframe, then decodes forward to the exact target time. Seeks are issued at millisecond precision, and the stamped labels adapt to the window — `m:ss` normally, sub-second (`1:07.3`) when a zoom window is short — with a hard floor of 2 ms per cell so adjacent cells always carry distinct addresses. Crucially, timecodes are **absolute to the source video at every zoom depth**: a value stamped on any sheet, however deep the zoom, refers to that same moment in the original file. ## Do my sheets or videos leave my machine? Depends which Squish you're using — the honest answer is a split, not a blanket "no": * **CLI / MCP server (this repo):** nothing leaves your machine, ever. Frames are extracted and composed locally; sheets are written to your disk. * **Web app (**[**getsquish.app**](https://getsquish.app)**):** processing runs in your browser; videos and photos are not uploaded. * **Hosted API (`api.getsquish.app`):** the one intentional upload — you send the video to get sheets back over temporary links. Full details, including what the hosted API retains and for how long: [privacy and data flow](/squish/the-primitive/privacy-and-data-flow.md). --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://getsquish.gitbook.io/squish/resources/faq.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.