Launched this week
Local-first meeting recorder for Mac — and fully open source (Apache 2.0). Records both sides of a call (mic + system audio) with no bot joining, transcribes on-device with Whisper, and summarises with your AI (Anthropic, OpenAI, Apple Intelligence). Exposes a local MCP server so Claude & Cursor can query your meetings — and doubles as push-to-talk dictation in any app. No cloud, no account, no subscription.
Hi everyone,
Daisy started as an internal tool. I do a lot of client interviews and discovery calls — design strategy work where the texture of what someone says matters as much as the substance — and for years my options were two:
1. Send the audio to Granola / Otter / Fathom / Shadow. Get fast summaries, lose control of where the transcript lives.
2. Type notes by hand and miss half the conversation.
I tried every cloud meeting tool. They're great products, but they all require a vendor relationship I didn't want for client conversations. The transcript sits on a server I don't own. Every new feature adds a subprocessor to their list. And the moment I asked "where exactly does my Acme client meeting live?", the answer was always "on our infrastructure, encrypted at rest, you can trust us."
I'd rather not have to trust anyone.
So I built Daisy. It records on the Mac — microphone plus the other side of the call captured through ScreenCaptureKit, no bot in the meeting. Whisper Large v3 transcribes on the Neural Engine. For summaries, you bring your own AI provider key — Anthropic, OpenAI, Apple Intelligence on macOS 26, or a local model via Ollama or LM Studio — and the key stays in your Keychain, with requests going straight from your Mac to the provider, never through our servers. The transcript lands in your folder of choice (mine is an Obsidian vault).
The structural thing competitors can't ship: Daisy exposes a local MCP server on 127.0.0.1. Claude Desktop and Cursor can query your transcripts directly — "summarise everything I discussed with Acme this quarter" — and Daisy answers from your own disk. No copy-paste, no upload, no API token sitting in someone else's database.
It's open source on GitHub under Apache 2.0 — OSI-approved, no commercial carve-outs. Built native on macOS, signed and notarised, with Sparkle auto-updates. Free during beta, no account, no subscription gate, no telemetry.
I'd love to hear:
— What's your current meeting-capture stack, and what would have to change for you to try Daisy?
— For developers: what's the ideal local-MCP query you'd want to run?
— For privacy-aware folks: any missing assurance you'd like me to address?
I read every comment. Thanks for taking a look.
— Egor (Addicted Studio)
RiteKit Company Logo API
@ca33u This is a refreshing approach to a real problem. The control-and-privacy angle resonates, especially for sensitive client work where you can't just hand transcripts to a third-party vendor. The fact that you're letting users choose their own AI provider is smart—it sidesteps the subprocessor creep you mentioned and keeps people from being locked into one platform's capabilities.
Building a fully local transcription pipeline with your own AI keys and an MCP server is a genuinely thoughtful approach to the data-sovereignty problem most meeting tools ignore. While Daisy solves the privacy side of capturing conversations, MentionFox's audio mention detection scans public video content and podcasts to surface moments where your product or competitors are being discussed — so you can find unsolicited user feedback and potential leads without anyone needing to trust a third-party server with private calls. If you're curious whether Daisy is already generating buzz in developer or privacy-focused communities, that's exactly the kind of signal MentionFox pulls automatically. Check https://mentionfox.com — in 30 seconds you can see where your product is being mentioned across text and audio sources right now.
@saulfleischman Thanks, Saul — appreciate you digging into the actual architecture, not just reacting to the pitch. The "don't lock users into one platform's AI" point is exactly the bet.
And fair plug — I'll check out MentionFox. Finding where something's discussed without touching private calls is a real gap.