MeetCap turns your coding agent into a meeting recorder, with local transcription
The first public Mac release pairs on-device transcription with Markdown meeting records and a skill for agents such as Claude Code, Codex, OpenCode and Pi.
Your coding agent can now have a meeting library to work with. MeetCap’s first public 1.0 release, published on October 18 at 11:44 UTC, combines a Mac meeting recorder with local transcription, Markdown records and an agent skill. Its developer says you can ask an existing agent to record a meeting, summarize what you missed or find a decision from an earlier conversation.
An October 19 Hacker News submission brought the project to our attention. That is a discovery signal, not evidence of broad adoption or the submitter’s claim that it is the most powerful note-taking app. The concrete news is the released workflow and its English setup documentation.
Your agent in, meeting records out
MeetCap is a standalone Mac application with a command-line tool and a skill, rather than a new language model or a fork of a coding assistant. Its guide lists Claude Code, Codex, OpenCode, Pi and other compatible agents. Installation adds the skill to supported agents it finds on the computer; an already-open agent conversation needs to be restarted to load it.
The documented requests include starting and stopping recordings, transcribing an existing file, writing minutes, naming a speaker, finding a past decision and exporting a meeting as a Word document. The agent guide says recording starts or stops when requested, and destructive changes such as deleting meetings or forgetting a voice require agreement first. Those are documented controls, not an independent audit of every agent’s behavior.
The developer describes live summaries during a meeting, a second transcription pass afterward, and minutes whose points link back to transcript lines and audio playback. English, Chinese and mixed-language transcription are advertised. AITrending has not reproduced recognition accuracy, speaker identification or summary quality.
Local audio does not always mean local AI
Transcripts and minutes live as Markdown files in a library the user chooses. The privacy notice says recordings stay on the computer that recorded them, while a selected remote AI service receives transcript text, meeting background and glossary context to produce summaries or minutes. An agent with access to the library can also read its contents under that agent provider’s terms.
Users can choose an existing agent, their own API key or a local model that MeetCap downloads. The application is offered free without an account or recording time limit; remote model charges and subscription allowances still depend on the provider. Choosing a local model keeps the text processing on the device, according to the project. Cloud-syncing the library is a separate user choice and can also sync saved voiceprints.
Mac first, proprietary app
Version 1.0 requires macOS 14.2 or later, on Apple silicon or Intel. Windows and iPhone versions are described as coming soon. The English guide offers a Mac download or a Terminal installer and installs the meetcap command alongside the app.
Microphone permission records your side; system-audio permission captures the computer’s output. If either is missing, the guide says MeetCap records only the available side. The terms require users to obtain applicable recording and voiceprint consent.
The app itself is not open source, despite its public GitHub release repository. The accompanying skill and reference files use MIT-0. For people already working through an agent, the useful proposition is straightforward: a local meeting archive that the same assistant can record, write up and search, with a choice over where the summarizing model runs.