AI Assist

AI Assist is a meeting companion that provides live transcription, AI summaries, and document-grounded Q&A during video calls. It runs as a Chrome extension paired with a desktop app, capturing meeting content directly from your browser tab without sending a bot to join the call. That means no third-party notification appears to other participants, no awkward recording disclaimer, and the audio streams from your machine straight to your own OpenAI account for transcription rather than passing through AI Assist's servers. The architecture is designed around privacy, making it a fit for teams with strict IT policies, regulated industries, or anyone uncomfortable handing meeting recordings to a SaaS provider that stores and processes them centrally.

Real-time captions are powered by OpenAI's Whisper API. Audio processing happens locally on your machine before being sent to Whisper for transcription, so the raw audio never touches AI Assist's infrastructure. You see captions appear as the meeting progresses, which is useful if you're in a noisy environment or joining a call where speakers have unfamiliar accents. The transcription becomes the foundation for everything else the tool does, including post-meeting summaries and the Q&A feature.

During meetings, the tool can surface relevant excerpts from files you've indexed on your device. If a topic comes up that relates to a document in your local folders, it displays the passage with a citation, no manual triggering required. This happens automatically in the background, though you can toggle it off if the proactive suggestions distract you. You can also ask questions mid-meeting and get answers grounded in your own documents. The important detail here is that only the matched text excerpts are sent to the AI model, while full files stay on your disk. That means you can index confidential documents without worrying that the entire contents will be uploaded to a third-party service.

The bring-your-own-keys model is central to how pricing works. You connect your own Anthropic and OpenAI API accounts and pay those providers directly at their standard rates. AI Assist doesn't mark up API usage, charge per seat, or lock you into an annual contract. The company estimates typical meeting costs at 50 cents to two dollars for LLM processing plus about 36 cents per hour for Whisper transcription, which they claim saves roughly 225 dollars a year compared to Otter's business tier at 19.99 dollars per user per month. The tradeoff is that you manage API keys, usage limits, and billing yourself rather than getting a flat-rate SaaS subscription with support and SLAs. If you already have an Anthropic or OpenAI account for other projects, adding AI Assist is lightweight since you're reusing credentials you already control.

The desktop license is a three-dollar one-time purchase, which is unusually cheap for productivity software. There's an optional cloud sync add-on for three dollars a month if you want end-to-end encrypted transcript syncing across devices, but without that the app works fully offline. A 14-day free trial gives access to full desktop features before you decide whether to buy. Currently supported platforms include macOS and Linux for the desktop app, with Google Meet and Zoom Web working through the extension. Windows and ChromeOS support is in development, as is Microsoft Teams Web compatibility. Mobile apps aren't on the roadmap, which limits flexibility if you take calls from your phone regularly.

The target user is a solo practitioner or small team that wants meeting notes without a SaaS subscription, without a bot joining their calls and announcing itself, and without sending recordings to a centralized service. Consultants, lawyers, therapists, and anyone else handling sensitive conversations will appreciate that the data path stays under their control. Setup takes about five minutes and requires your two API keys plus a free AI Assist account for pairing and optional cloud features. The indexed-file Q&A turns it into more than a transcription tool, giving you a way to pull relevant context from your own knowledge base into conversations as they happen.

Where it falls short is platform coverage. Windows users are still waiting, Teams support isn't ready, and there's no mobile option. If you're in a Microsoft-centric organization or need transcription on the go, this isn't the right tool yet. The BYOK model also means you need to be comfortable managing API credentials and understanding usage costs, which is a barrier for less technical users. But for macOS and Linux users on Meet or Zoom who value owning their data and controlling costs directly, AI Assist hits a privacy-first niche that the major meeting transcription players ignore entirely.