Google’s new offline note-taker takes aim at Granola with on-device AI

Google has released a new Mac meeting-notes app called Google AI Edge Foresight that can run entirely offline, positioning it as a direct competitor to AI note-takers such as Granola. The app matters because it brings Google’s on-device AI push into one of the most competitive productivity categories, where privacy, speed and reliability are increasingly important.

The company’s latest release builds on an earlier April dictation tool that also relied on local models. This time, however, Google is going further: the app is designed to transcribe conversations, generate structured notes, answer questions from meeting content, and work without any internet connection at all.

What Google AI Edge Foresight does

Google AI Edge Foresight is a Mac application built for taking notes during meetings, whether those meetings happen on video calls or in person. The app uses Google’s on-device EmbeddingGemma 2 model, which has 740 million parameters, to process speech and help turn rough notes into more usable meeting summaries.

The product’s interface will look familiar to anyone who has used Granola. Google uses a split-screen design that puts shorthand notes on one side and AI-generated output on the other, making it easy to compare a user’s raw input with the assistant’s expanded version.

Users can also keep typing manual notes while the app is transcribing in the background. That feature is important because many people do not want a fully automated system that replaces their own workflow; they want something that augments it.

How is it different from a basic dictation app?

It is more than dictation because it does not just convert speech to text. Google AI Edge Foresight also organizes information, answers questions about the meeting, and lets users search or chat with the assistant after the session ends.

In Google’s setup, the transcript remains available for review, and the app includes a Gemma 4-powered assistant that can respond to questions about what was said. That makes the tool closer to a meeting memory system than a simple transcription utility.

Why offline AI is the big selling point

The most notable feature of Google AI Edge Foresight is that it can operate fully offline. According to Google’s FAQ, the app is optimized for Apple Silicon and does not require a live internet connection to function.

That design has several practical benefits. It can be used while traveling, in low-connectivity environments or in workplaces where teams do not want meeting data sent to the cloud. It also reflects a broader industry shift toward local-first AI tools that keep sensitive work on the device.

For a meeting assistant, offline operation is not just a technical flourish. It can reduce latency, improve reliability and offer a stronger privacy story than cloud-dependent products. In a market where professionals are increasingly cautious about where their notes go, those advantages can be decisive.

Google’s positioning suggests the product is meant to show what its Gemma family can do on a laptop without needing a server connection, while also leaving room for a more consumer-facing version later.

How does Google AI Edge Foresight compare with Granola?

It is similar to Granola in both layout and purpose, but Google is using its own offline AI stack to differentiate the product. Granola helped popularize the modern AI notetaker format by pairing a user’s rough notes with polished AI output, and Google is clearly borrowing that core experience.

Where Google may have an edge is in its local model strategy. Rather than depending entirely on cloud services, the app uses on-device AI to do the heavy lifting. That could appeal to users who want better confidentiality or who simply prefer tools that continue working even when Wi-Fi does not.

Granola remains one of the best-known names in this space, but Google’s entry adds weight to a category that is already getting crowded with ambitious new products.

What can the app read and remember?

Google AI Edge Foresight is designed to work with more than spoken conversation. Users can upload a range of documents to build a personal knowledge base that the app can draw from during meetings and later when answering questions.

Supported file types include:

• PDFs
• Google Docs
• Microsoft Office files
• Plain text
• Markdown
• Web bookmarks

That document support turns the app into a broader research and memory assistant. If a subject arises during a meeting that connects to something in the uploaded materials, the assistant can surface relevant information in real time.

In practice, that means the app is not only summarizing what happened in the room. It is also trying to provide context from the user’s own files, which could make it especially attractive to consultants, managers, researchers and students who move constantly between live discussion and stored reference material.

What is Google really trying to prove?

Google’s release may be partly about showcasing what its local AI models can do. The company has been steadily expanding its Gemma family, and this app is a practical demonstration of how those models behave in an everyday productivity setting.

That does not necessarily mean the company is stopping at a demo. Google could eventually offer a more mainstream version that works across video-calling platforms through its Gemini apps and uses cloud models for broader functionality. In other words, the current Mac app may be both a product and a prototype.

This dual purpose would make sense. Tech companies often use limited releases to test product ideas, gather user feedback and refine model performance before they commit to a wider launch. Google’s new note-taker appears to fit that pattern.

Why the meeting-notes market is getting crowded

The race to automate meetings has intensified as AI assistants have become easier to build and more useful to everyday workers. What was once a niche category for startups has quickly become a mainstream productivity battleground.

Over the past few months, dictation company Wispr and scheduling platform Calendly have each introduced their own note-taking products. In another sign of consolidation, productivity company Superhuman, which was formerly known for Grammarly after a broader rebrand in market perception, acquired YC-backed note-taker Fathom last month.

That activity shows two things. First, meeting intelligence is now seen as a core workplace feature rather than a novelty. Second, smaller specialists are being squeezed by larger players that already have distribution, brand recognition and product ecosystems in place.

Who is competing for the same user?

Everyone from startup founders to established software companies wants a share of the same workflow: capture the meeting, summarize the meeting, and make the notes searchable afterward.

That workflow has clear commercial appeal because it sits at the intersection of note-taking, transcription, knowledge management and conversational AI. It is also sticky. Once teams adopt a tool for meetings, they often begin storing more and more institutional memory inside it.

How local-first AI changes the product story

Local-first design changes both the technical architecture and the user promise. Instead of sending raw meeting audio to remote servers first, the app performs more of the work directly on the device. That can improve responsiveness and reduce exposure of private content.

For Google, local-first AI also helps demonstrate that modern models are not limited to cloud-based applications. A 740 million parameter model running on Apple Silicon is a strong example of on-device capability, especially for users who have assumed advanced AI must be hosted remotely.

There is also a strategic angle. If Google can deliver strong note-taking performance offline, it may be able to compete on trust as well as on features. In productivity software, trust can matter just as much as raw intelligence.

Where the product could go next

The current release is available as a Mac app, but the broader opportunity is larger than a single desktop tool. Google could fold similar capabilities into its Gemini products, bring the feature to more platforms, or build a more polished consumer-facing assistant around the same core technology.

That path would align with the company’s broader AI strategy: develop advanced models, prove they work in a constrained setting, then distribute them through the products people already use. Meeting notes may seem narrow, but they are often the entry point for broader workflow automation.

If Google can make the experience seamless enough, the app could evolve from a model showcase into a habit-forming productivity product. The challenge, as always, is proving that the assistant saves time without creating new friction.

Key facts at a glance

Timeline of Google’s note-taking push

Bottom line

Google’s new offline note-taker is more than another productivity app. It is a signal that local AI is becoming good enough to power real-world meeting workflows, and that major tech companies now see note-taking as a serious front in the race for workplace AI.

By combining offline processing, a familiar Granola-style interface, document-aware search and question answering, Google has entered a crowded but commercially promising market with a product that is as much a technology statement as it is a consumer tool.