LayerNorm
Personal computing interface company
✽
LayerNorm is a software company founded in February 2026 by Divyansh Lalwani, a Biomedical Engineering graduate of The Johns Hopkins University.7 The company is based in Baltimore, Maryland, and builds human-computer interfaces for AI.71 Its main product, Overlay, is an open-source AI workspace that aggregates models from multiple AI labs — including OpenAI, Anthropic, Google, and xAI — into a single interface starting at $8 per month.4 The company's name references layer normalization, a mathematical operation used in transformer neural networks.
Lalwani began building the software that would lead to LayerNorm while studying at Johns Hopkins. In June 2025, while still a student, he launched AutoQuill — a macOS voice dictation application built with the Flutter framework and powered by Groq's inference API running the Whisper speech recognition model.96 Lalwani graduated from JHU in December 2025 and formally incorporated LayerNorm in February 2026.78
The founding thesis, stated in Overlay's published manifesto, is that AI models are becoming commodities while the interface layer between humans and those models is the locus of lock-in.3 The manifesto argues that AI tools in 2026 fragment user context across separate subscriptions, applications, and data silos, and that an open, extensible aggregator can gain the interface layer by the same pattern that saw Chrome gain market share from Internet Explorer and VS Code become more widely adopted than JetBrains-style editors.3 Lalwani spent the summer of 2025 in San Francisco before founding the company, describing that period as formative for his decision to pursue software entrepreneurship full-time.8
AutoQuill is a macOS application that transcribes voice to text system-wide using a global hotkey, working inside any application.5 It was built using the Flutter framework and released under the MIT license, with source code available on GitHub.6 The application calls Groq's API to run the Whisper model, achieving sub-500-millisecond response times.5
The app offers three transcription modes. Standard mode starts and stops recording on a single hotkey press, automatically copying the result to the clipboard. Push-to-talk mode records while a key is held down and transcribes on release. The AI assistant mode allows users to select text in any application, then speak instructions such as "make this more formal" to produce an edited version.6 Users supply their own Groq API keys; no data is transmitted to LayerNorm's servers.5
AutoQuill's free-forever positioning distinguishes it from competitors in the voice dictation category. WisprFlow's free tier limits users to 2,000 words per week; SuperWhisper restricts free users to small models; AutoQuill imposes no word caps, supports custom dictionaries, multilingual input, and configurable hotkeys at no cost, and allows users to bring their own API keys.5
Overlay is an open-source AI workspace available at getoverlay.io. It consolidates chat, notes, projects, knowledge bases, and automations into a single interface, with shared context that persists across sessions so files, notes, and project artifacts remain linked between conversations.2 Lalwani compares Overlay to what the web browser was for the internet: a platform developers can extend with their own applications, rather than a walled garden controlled by a single AI laboratory.3
Overlay supports models from multiple providers simultaneously — OpenAI's GPT models, Anthropic's Claude, Google's Gemini, xAI's Grok, and DeepSeek, among others — allowing users to select the model best suited to each task rather than committing to a single vendor.2
The product originally launched as a macOS desktop application with four overlay panels — speech transcription, a scratch-pad for notes, an AI chat window, and an embedded browser — designed to appear on top of the user's current screen and disappear on demand. Lalwani described the design in a March 2026 interview: "They appear when you want them there, and you can just hide away when you don't want them there."8 The web version at getoverlay.io, launched later in 2026, requires no installation and offers unlimited use of an auto-selected model on the free tier.10
Overlay uses a tiered subscription model. Each budget unit represents $1 of model API spend. The free tier provides unlimited access to an auto-selected model, notes, chats, knowledge bases, and project management, with 10 MB of file storage.4 The Starter plan ($8/month) includes 8 budget units, 1 GB storage, access to premium models, browser task automation, image and video generation, and code execution via Daytona sandboxes. Pro ($24/month) provides 24 budget units and 3 GB storage; Max ($94/month) provides 94 budget units and 11.8 GB storage.4 Users can also set a custom monthly budget between $8 and $200. One-time top-ups of $8 are available on demand; automatic top-ups trigger when a plan's budget reaches zero, but are disabled by default and require explicit opt-in.4 An enterprise tier with a contact-sales model is available for organizations that require additional controls.2
The web client source code for Overlay is published on GitHub under the repository DevelopedByDev/overlay-web.2 Lalwani's manifesto argues open source is a strategic necessity, not only an ethical position, citing the consistent historical pattern of open platforms winning market share from closed ones.3 The manifesto notes that Android and Linux collectively run more devices than macOS, Windows, and iOS combined, and predicts the same consolidation around open standards will occur in AI interfaces.3