Transcribe, summarize, and search everything you've heard — locally, in minutes.
Your machine. Your models. Your knowledge base.
$ brew install --cask audistill/tap/audistillFrom raw audio to searchable knowledge — entirely on your machine.
Files, YouTube links, RSS feeds, or any URL. One input field auto-detects everything.
30–50× realtime on Apple Silicon
Transcribed on-device in minutes using Parakeet TDT on Apple Silicon. Then summarized, structured, and tagged by the LLM of your choice.
Audio never leaves your machine.
Search across everything. Ask questions. Surface patterns. Generate new content from your knowledge base.
Every tier. Every feature. No upsells.
A research assistant with tools. Search across your library, extract patterns, and generate new content — not just a chatbot.
Find every mention of churn across last month's interviews
3 matches found
Files, YouTube links, RSS feeds, or any URL. One field, auto-detected.
30-50x realtime on Apple Silicon. Audio never leaves your machine.
Use any LLM with your own API key. No markup, no middleman.
Define how content gets shaped. We call them Recipes — run on any transcript.
Find anything across every transcript and every generated document.
Same app. Different workflows. Same result: knowledge you can actually find again.
Paste YouTube links from WWDC, JSConf, Strange Loop. Build a searchable library of every talk you watched.
…RSC eliminates the waterfall by moving data fetching to the server…
…server components change how we think about the component tree…
…the mental model shift with RSC is fundamentally about ownership…
↑ found across 6 months of talks
Drop in the recording. Get structured decisions and action items in seconds.
Subscribe to feeds. Auto-ingest. Search across months of listening.
Ingest 15 customer calls. Ask one question across all of them. See what actually keeps coming up — not what you assumed.
Run different Recipes on the same 90-minute recording. Brief for review, detailed for study, key quotes for your paper.
The lecture introduces three competing models of attention allocation in cognitive load theory, with emphasis on the redundancy effect.
Key distinction: intrinsic vs. extraneous load — the speaker argues most instructional design failures are extraneous.
Actionable takeaway: chunk information into 3–5 segments before testing recall.
Every tier includes all features. Forever.
The only difference is how many Macs you own.
No. Transcription runs entirely on your Mac using Apple Silicon. Audio files never leave your machine. AI features (summaries, chat) use your own API key — calls go directly from your Mac to the provider you choose.
You provide your own API key (e.g., OpenRouter, Anthropic, OpenAI). Audistill sends your transcript to the model you pick for summaries and chat. Typical cost: a few cents per episode. We never see your key or your data.
The app remains viewable — you can browse your library, read transcripts, and search. Ingest, chat, and recipe execution require a license to continue.
“Your audio never leaves your machine. Transcription happens entirely on-device.”
No. One-time purchase, lifetime updates. Pick the tier that matches how many Macs you own.
Any Mac with Apple Silicon (M1 or later). macOS 13 Ventura or newer. Intel Macs are not supported.
The source code is available. You can build and run it yourself. The paid download gives you a signed, notarized binary with auto-updates and supports continued development.
Audistill is fully open source. Inspect, modify, contribute —
your tool, your rules.
Main repository
14 days free. No credit card. Pick your path.
Signed & notarized .dmg
One-click install. Auto-updates. Supports development.
Download for MacFor terminal lovers
Install and update via Homebrew Cask. Same signed binary.
$ brew install --cask audistill/tap/audistillFull control
Clone, inspect, build. Free forever. No telemetry, no limits.
View on GitHub