Granola, an AI-driven notetaking application with a valuation of $250 million, has gained significant traction among technology founders and venture capitalists. However, a developer identified a market need for a more private, entirely local alternative, offered through a one one-time purchase rather than a subscription model. This insight culminated in the development of a new Mac application named Talat.
Nick Payne, a developer based in Yorkshire, England, who describes himself as a "computer nerd," attributes the genesis of his local AI notetaker project largely to a sequence of serendipitous discoveries.
"Granola is, in my opinion, fantastic; it stands as a brilliant illustration of the potential of an Electron app [a desktop application framework] when developed with dedication," Payne shared with TechCrunch. He added, "My initial experience with it left me intrigued by its ability to capture system audio on my Mac independently of video, which was the prevailing method then. This fascination prompted extensive research, ultimately uncovering a relatively novel and insufficiently documented Apple API."
To simplify interaction with this specific API, known as Core Audio Taps — which allows developers to access a Mac's audio streams — Payne opted to develop an open-source audio library called AudioTee.
Payne explained, "Throughout that period, I was progressively assembling a suite of tools, yet nothing emerged that truly felt viable as a standalone product, rather than merely an impressive technological demonstration." He continued, "The cutting-edge hosted transcription models, utilized by services such as Granola, are remarkably advanced, and witnessing your speech materialize on screen in near real-time is genuinely captivating. However, it consistently bothered me that this convenience necessitated surrendering not just my general data, but specifically my audio data — my actual voice."
His breakthrough came with the discovery of FluidAudio, a Swift framework designed for fully local, low-latency audio AI processing on Apple hardware. This toolkit facilitates the direct execution of compact, high-speed transcription models on the Mac’s Neural Engine, Apple’s specialized AI processing unit.
This crucial finding solidified Payne's vision, enabling him to transform his extensive research into a tangible product where user audio remains exclusively on their Mac, and transcripts are never stored on external company servers.
Talat, a culmination of Payne's deep interest in audio technology, was developed in collaboration with his long-standing friend and former colleague, Mike Franklin. The application is a compact 20 MB download, offered as a one-time purchase, eliminating the need for account creation or sharing of analytics data with the developers. Furthermore, it is entirely free of recurring charges.
Although some AI notetaking solutions boast a broader array of functionalities, Talat provides a focused and efficient feature set. It records audio via your computer's microphone during virtual meetings held on platforms such as Zoom, Teams, and Google Meet, transcribing it instantaneously. The application attempts real-time speaker identification, with provisions for manual reassignment. Users can also add personal notes, and modify, remove, or segment transcript entries. Upon meeting conclusion, an integrated local LLM automatically produces a summary, highlighting essential points, decisions, and actionable items.
All notes, transcripts, and summaries within Talat are fully searchable.
Beyond its privacy-centric approach, Payne emphasized that a core objective is to empower users with greater control and choice.
He elaborated, "We are prioritizing configurability, granting users authority over their data's destination: this includes selecting their preferred LLM, enabling automatic export to Obsidian [a note-taking application], utilizing webhooks to transmit data upon meeting completion, and integrating an MCP server — a standardized mechanism for AI tools to interface with external data sources — for on-demand data retrieval."
Internally, Talat's AI architecture is a composite, "primarily assembled and abstracted through FluidAudio," Payne stated, acknowledging its significant contribution to the system's core operations. For summarization tasks, the application defaults to the Qwen3-4B-4bit AI model, which is designed to operate effectively even on less powerful hardware.
Users, however, retain the flexibility to interchange this with any cloud LLM provider they prefer, select from two Parakeet variants — Nvidia's speech recognition models — or direct it to Ollama, a platform for local AI model execution, thereby enhancing their control over the application. Future updates for Talat are planned to introduce additional integrated options and expand integrations with other applications, including Google Calendar and Notion.
Upon its release, individuals utilizing M-series Mac computers (equipped with Apple's proprietary processors, beginning with the M1 chip) are eligible to download and experience the app free of charge, with a trial period allowing up to 10 hours of recordings before committing to a purchase.
The pre-release version of Talat, which remains under active development, is currently available for $49.
Upon reaching its official 1.0 release, the price of the application is slated to increase to $99.
Payne and Franklin are independently funding Talat's development and intend to maintain the core product as a one-time purchase model for the foreseeable future.
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.