an Affordable Magic product
Point synop.stream at a YouTube channel, meeting page, or podcast feed. It produces editor-grade synopses — TLDR, key moments, decisions, quotes, comment digests — and publishes them to a subdomain you own. Local-first — runs entirely on your hardware.
For people who cover a beat — and want the synopsis, not the three-hour video.
Local-first — your model, your hardware, your data.
Closed beta
The host sits down with a veteran indie founder to unpack a decade of building in public — what compounding actually felt like, why they turned down funding twice, and the one metric they wish they’d ignored sooner.
Votes with roll calls. Quotes with attribution. Public comment digests. The synopsis takes a shape that matches your beat — not a generic three-bullet summary.
Plain-English criteria: “interviews and full episodes — skip trailers and shorts.” Quick checks first, full reads only when ambiguous. Everything rejected stays visible with the model’s reason — so nothing goes missing quietly.
Drop in your board members, acronyms, and district names. The model uses them instead of hedging or generalizing — so the synopsis reads like someone who covers the beat wrote it.
Raw transcript, full chat, complete comment thread — preserved alongside every synopsis. Verify a quote, check the moment, or share the original without leaving the page.
synop.stream runs end-to-end on your machine. Transcription happens on your chip, the LLM runs in your own Ollama, and your media never touches a third-party cloud. The only bytes that leave are the static site you choose to deploy — and only to the Cloudflare account you control.
Set it up once. Let the scheduler do the rest.
A YouTube channel, a page of meeting minutes, or both. Add once; the scheduler watches from then on.
Cheap metadata pass first. Full-content re-check only when the metadata is ambiguous. Everything rejected stays visible in an audit view, with the reason.
Template plus your reference context drive the main summary. Viewer comments and live chat get their own dedicated summaries. Raw originals stay one click away.
Static site built per project. Cloudflare Pages deploy, custom domain, DNS CNAME — all provisioned on first run.
A listener creates a project called “The Long Take,” pastes the show’s YouTube channel (or RSS feed) URL, and writes a short criteria line: “Full episodes and interviews. Skip trailers and clip reels.” They add reference info — the hosts’ names, recurring guests, a few acronyms the show leans on. They pick the Podcast template and toggle Auto-process + Auto-deploy.
Three weeks later, a dozen episode synopses live on their subdomain. Each opens with a TLDR, then chapters with timestamps, the notable quotes with attribution, a digest of listener comments, and the raw transcript one click away. They didn’t touch a button. The ones that didn’t fit — a trailer, a best-of rerun — landed in Filtered Out with the model’s reason attached, so nothing went missing quietly.
A city commission, a county board, a podcast, a research topic tracked across sources — each a different format and a different shape of synopsis. Same engine, same templates, different beats.
Apple Silicon Mac (M1 or newer). Larger models run more comfortably on Pro/Max/Ultra chips. We’ll share recommendations during onboarding so you pick the model that fits your machine.
A free runtime that lets you run open-weight LLMs locally on your Mac. Install once, pick a model, you’re set. We walk you through it during onboarding. ollama.com
Only for the auto-deploy step. Synopses are produced locally; the built static site can be hosted anywhere — or read directly from your own machine.
Not yet. Apple Silicon’s local-LLM tooling is the most polished today. Linux is on the roadmap; Windows further out.
Onboarding is rolling and small. Expect a real conversation, not an autoresponder.
It’s already yours — every transcript, synopsis, and audit log lives on your hard drive in plain files. No lock-in by design.
We’re onboarding a small group of early users. Best fits so far: city and county coverage, beat reporting archives, organization-internal digests, researchers tracking a topic across sources. If you have a clear use case and a Mac to run it on, we’d like to hear from you.
Write to usOr write directly to [email protected]
Built by an indie developer who wanted a slow, boring feed of local facts instead of a fast, angry feed of national ones.