Local-first meeting capture with agent access for privacy-focused teams
doodle-note, by Onyx Dev Labs, is a local-first meeting capture and AI note-taking app designed to run on a user’s machine. It records meetings, transcribes speech, and generates summaries using models hosted locally or supplied by the user. The app supports calendar-aware recording prompts and searchable transcripts, making it suitable for professionals and researchers who need private, machine-resident meeting records and AI-driven retrieval.
What tasks can you actually use doodle-note for?
The app targets meeting capture, transcription, and summary generation without adding external bots. It records from the microphone and system audio so calls do not require a joined bot, and it produces searchable transcripts and AI-generated notes that serve as meeting documentation. For users who need timestamped meeting logs or condensed meeting notes, the app converts live audio into text and extractive summaries directly on the desktop.
How reliable are its local transcriptions and summaries?
Transcription runs on on-device speech engines, specifically Swift and FluidAudio on Mac and a local engine on Windows, so output quality depends on those engines. Summarization uses models that can be run locally or supplied by the user, therefore the fidelity of generated notes reflects the chosen model and its configuration rather than a single cloud service. Expect variation in output when switching models or engine settings.
What inputs and environment does it require to work well?
The app targets Apple Silicon macOS 14 or later and WindowsBeta), and it integrates with the user calendar to trigger recording start and stop prompts. It maintains a searchable history across past meetings and transcripts, which supports later lookups. Because the Windows release is described as Beta, cross-platform parity and stability can differ between the two desktop environments.
Does it integrate with AI agents while keeping data local?
doodle-note exposes a read-only Model Context Protocol server so MCP-enabled agents can query local meeting records without remote upload. The design keeps processing on the device and provides agent-side search access to transcripts, enabling AI-assisted retrieval while avoiding external audio transfer. Compatibility depends on the client supporting MCP and on the user opting into the read-only server.
Practical choice for professionals who prioritize local control and agent access
doodle-note suits developers, researchers, and privacy-conscious professionals who need meeting records to stay on their machines and want AI agents to reference that archive. It requires configuring local or BYO models for consistent summaries and offers limited Windows stability during Beta. For teams prioritizing on-device control and agent-enabled search, it is a focused, practical option.
