Capture every mic and the system audio on separate tracks, let a local model strip the bleed, get an accurate transcript back, edit the episode by editing the text, and draft the show notes, chapters, and newsletter — all on your Apple Silicon GPU through local UVR, Whisper, and language models. No upload, no account, no cloud. A studio for shows that can't leave the room.
Silo Systems builds infrastructure for people who would rather not negotiate with their own software. SiloCast is the same discipline, pointed at one job — turning a room full of microphones into a finished, published episode — done entirely on the machine you already own. Your unreleased show never becomes someone else's training data, and your back catalogue never lives on someone else's server.
Recording, bleed separation, transcription, and content generation all run on your Mac through a bundled local runtime on your Apple Silicon GPU. The separation, speech, and language model weights download once from public open-weights mirrors, then SiloCast works offline for the lifetime of the app — in a locked studio exactly the way it works at your desk.
No analytics SDK, no telemetry, no account, no crash reporter, no external API. The Python inference layer binds strictly to loopback (127.0.0.1:8765) as on-device IPC — it is not a network service. The audio you record, the tracks SiloCast isolates, and the copy it drafts stay in your Application Support container, never transmitted anywhere.
Show notes, episode outlines, timestamped chapters, and newsletters are generated locally by a language model through mlx-lm. The post-production a cloud service would charge you a subscription for — and feed your unreleased episode into — is computed on hardware you already own and kept there.
SiloCast records each USB or XLR microphone and the system audio loopback simultaneously through Core Audio, each captured to its own track. Whoever is in the room and whoever is on the call land in the same session — no virtual cable, no second recorder, no bot joining your show.
A local UVR model (HT-Demucs) isolates each voice and strips the bleed, so the host mic no longer carries the guest and every track edits independently. Then a Metal-accelerated Whisper model (mlx-whisper) transcribes the cleaned audio on your GPU, attributed to the right speaker — entirely on-device.
Tighten the episode by deleting sentences in the transcript — the audio is cut to match. When it's done, a local language model through mlx-lm drafts show notes, an episode outline, timestamped chapters, and a newsletter. The whole marketing pack a cloud service would charge you for, computed on your own machine and kept there.
Record several USB or XLR microphones and the system audio loopback at the same time, each on its own track — so every host, every guest, and the call all land in one session, cleanly separated from the start.
An optimized local UVR model (HT-Demucs) isolates each voice onto its own track and strips the bleed, so the host mic stops carrying the guest and every track is clean enough to edit independently.
Multi-speaker speech-to-text runs through a Metal-accelerated Whisper model (mlx-whisper) on your Apple Silicon GPU. Fast, attributed to the right track, and entirely on-device.
Select a sentence in the transcript and delete it — the underlying audio is cut to match. Tighten an episode, drop a tangent, or remove a stumble without ever opening a waveform.
A local language model via mlx-lm turns the finished episode into show notes, an episode outline, timestamped chapters, and a newsletter draft — no cloud notetaker, no copy-paste into a chatbot.
Export isolated tracks, the mixed episode, the transcript, and every generated asset to a folder you choose. Your audio, your files, no re-upload and no watermark.
UVR, Whisper, and the language model download once from public open-weights mirrors with live byte-progress. After that, every separation, transcript, and draft is offline.
A SwiftUI host boots a bundled FastAPI backend on loopback and renders the editor in a WKWebView from that same origin. No browser, no server, no port open to your network.
Hosted recording studios and AI post-production tools rent you the workflow by the seat and run your unreleased episodes through their servers. SiloCast is a single purchase, on your Mac, with unlimited local recording, separation, transcription, and content generation — because it all runs on hardware you already own. Buy once, keep the tape.
Coming to the Mac App Store ⌘SiloCast is in pre-release for the Mac App Store. Join the waitlist and we'll send a single email when it goes live — no marketing list, no newsletter, no follow-up campaigns.
