AI meeting notetakers are having a moment, especially among founders and venture capital circles. Granola’s popularity and lofty valuation have helped spotlight just how useful real-time transcripts and instant meeting summaries can be. But not everyone is comfortable sending their voice recordings to remote servers or paying indefinitely for a subscription.
That gap is exactly what inspired Talat, a new local-first AI notetaking app for Mac built by Yorkshire-based developer Nick Payne alongside his longtime friend and former colleague, Mike Franklin. Their pitch is simple and timely: a lightweight, one-time purchase AI notetaker that keeps your audio on your Mac, doesn’t require an account, and doesn’t phone home with analytics.
How Talat was born: a privacy-first alternative to cloud AI transcription
Payne says the idea didn’t start as a business plan. It grew out of hands-on experimentation with macOS audio capture and a series of discoveries while investigating how modern meeting transcription apps record system audio without relying on clunky workarounds.
That curiosity led him into deep research around a newer, poorly documented Apple audio capability known as Core Audio Taps, which makes it possible to tap into audio streams on a Mac. Along the way, Payne built an open-source audio library called AudioTee to make working with that audio tech easier.
But the real turning point came from a different realization: cloud transcription is impressive, but it often comes with an uncomfortable tradeoff. Even when the experience feels magical, you’re still uploading highly sensitive data—especially your raw voice audio—to third-party services.
Local AI transcription on Mac: the key tech that made Talat possible
Talat became viable when Payne discovered FluidAudio, a Swift framework designed for fully local, low-latency audio AI on Apple devices. By running smaller, efficient transcription models directly on Apple’s Neural Engine, Talat can generate live transcripts without sending audio to the cloud.
That’s the foundation of Talat’s core promise: your meeting audio stays on your device, transcripts aren’t stored on someone else’s servers, and you can still get fast, practical AI notes.
What Talat does: real-time transcripts, speaker labels, searchable notes, and AI summaries
Talat focuses on a streamlined set of features aimed at people who spend a lot of time in online meetings and want accurate notes without sacrificing privacy.
It records audio from your Mac’s microphone during calls on popular meeting platforms like Zoom, Microsoft Teams, and Google Meet, then transcribes the conversation in real time. It also attempts to identify and assign speakers as you go, and if it gets it wrong, you can manually reassign speakers afterward.
During or after the meeting, you can take notes and edit the transcript by deleting, splitting, or cleaning up segments. Once the meeting ends, Talat uses a local large language model to generate a structured summary that highlights key points, decisions, and action items—so you don’t have to comb through a full transcript to figure out what matters.
Everything inside Talat is searchable, including transcripts, notes, and summaries, making it useful not just for capturing what happened, but for retrieving details later.
Configurability: choose your LLM, export your notes, and control where data goes
Beyond privacy, Talat is designed around user control. Payne says the team wants to give people options in how they manage meeting data and which AI models they rely on.
Users can configure Talat to:
– Pick their preferred LLM (local or cloud)
– Auto-export meeting notes to Obsidian
– Trigger webhooks when a meeting finishes
– Use an MCP server, a standardized method for AI tools to connect to external data sources and pull information on demand
This approach is meant to appeal to power users who want their meeting notes to flow into their existing knowledge base, automation stack, or workflow—without being locked into a single ecosystem.
Which AI models Talat uses (and how you can switch them)
Talat’s AI setup is a mix of components, with FluidAudio doing much of the heavy lifting behind the scenes. For summarization, Talat defaults to Qwen3-4B-4bit, a model lightweight enough to run on relatively modest Mac hardware.
If you prefer other options, Talat also allows you to:
– Switch to a cloud LLM provider of your choice
– Choose between two Parakeet speech recognition model variants developed by Nvidia
– Point the app at Ollama to run supported AI models locally
The developers say more built-in options and integrations are planned, including support for additional apps such as Google Calendar and Notion.
Pricing and availability: one-time purchase, no subscription
Talat is currently available for Macs with Apple silicon (M-series chips starting with M1). New users can try the app for free with up to 10 hours of recordings, giving them time to test real-world meetings before paying.
While Talat is in its pre-release stage and still under active development, it’s priced at $49 for a one-time purchase. When it reaches its 1.0 release, the price is expected to increase to $99.
Payne and Franklin are bootstrapping the project and say they intend to keep the core product a one-time purchase rather than moving to a subscription model.
Why Talat stands out in the AI meeting notes space
As AI note-taking apps become more common, Talat positions itself as the privacy-first choice for Mac users who want local transcription, local summaries, and control over integrations—without creating an account, paying monthly, or handing over voice recordings.
For anyone looking for a local AI meeting assistant on Mac, Talat’s combination of real-time transcription, searchable notes, and on-device summaries offers a compelling alternative to cloud-only notetakers.
