by andyhuo520
Free, open-source macOS app that transcribes recordings on your Mac, writes notes and to-dos, and exposes six MCP tools to Claude Code and Codex.
Choose BeanPod if you want private, on-device transcription on a Mac and a notes library your AI assistant can search. It can replace a cloud recorder subscription for one user. The verdict: promising, but a first release from one developer that needs macOS 26 and offers no support, so pilot it before relying on it.
It records, transcribes and summarises on the Mac, then serves the library to your AI assistant over a local MCP server.
Per the README, notes export as Markdown and recordings as M4A or WAV, so little locks a user in and the value is the local pipeline.
Main trade-off
A single-developer v0.1.0 release that needs macOS 26 and a manual security override to open.
Free and open source under the MIT licence, with no paid plan. Cloud transcription or cloud note models, if you turn them on, are billed by those providers against your own API key. Source: the BeanPod README and its GitHub release page.
BeanPod is a free, open-source native macOS app in the AI note-taking category. It records, transcribes and summarises audio on the Mac, pairs with the Anker soundcore D3200 recorder or works without it, and runs a Model Context Protocol server with six tools. The code is MIT licensed and the first release was v0.1.0.
On-device transcription
Qwen3-ASR 0.6B runs through llama.cpp with timestamps, the model download is checked by SHA-256, and macOS speech recognition is the fallback.
Notes and to-dos
Each recording is classed as a meeting, interview, sales call, phone call or lecture, then given a summary, key points, to-dos and people.
Knowledge base questions
You ask across all recordings and each answer names the recordings it relied on; filters cover time, scene and person.
MCP server and command line
Six tools, including search_recordings and add_note, are served to Claude Code, Codex, OpenClaw and Claude Desktop, with a local-only address at 127.0.0.1:8791.
Sentence replay and English study mode
Click any sentence to replay it or loop it, and for English audio the app lists 20 to 30 key words with level and an example sentence.
Many ways to bring audio in
Use the soundcore D3200 over Bluetooth or Wi-Fi, the Mac microphone and system audio, imported files, podcast and video links, or Feishu Minutes sync.
Voice dictation
Hold a key on the recorder and speak, and the text is typed into the app you are using, with a searchable history kept apart from recordings.
| Tool | Price from | Free plan | Runs on | Rating |
|---|---|---|---|---|
| BeanPod This page | Free | Yes | Mac | 3.1 |
| $17.99/mo | Yes | Web · iOS · Android | 3.5 | |
| $8.33/mo | Yes | Web · iOS · Android · Mac · Windows | 4.4 | |
| $4.99/mo | Yes | Web · iOS · Android | 4.3 |
MIT
Nothing. BeanPod is open source under the MIT licence and the README and release page list no paid plan. Your own costs are the Anker soundcore recorder if you want one, and any cloud transcription or note models you switch on, which the provider bills against your key.
Yes. The README says it runs as a recording and learning app on its own: record with the Mac microphone or system audio, import audio files, or paste podcast and video links. Transcription and notes work the same way. Only the device pages and Wi-Fi fast transfer need the soundcore D3200.
Claude Code, Codex, OpenClaw and Claude Desktop connect in one click from the Connect AI assistants menu. Any other MCP client can use the manual config or the local address http://127.0.0.1:8791/mcp, which exists only while BeanPod is open and listens on your own machine.
BeanPod is one. It transcribes with Qwen3-ASR 0.6B through llama.cpp, or macOS speech recognition if the model is not installed, and neither uploads audio. The model download is about 1 GB, and notes need a separate local model of about 2.5 GB or a cloud key.
Not by default. Recordings, transcripts and notes are kept in the BeanPod folder under Application Support on your Mac. Audio goes out only if you choose cloud transcription or ask for a single recording to be retranscribed in the cloud, and transcripts go out only if you configure a cloud note model.
BeanPod (豆仓 in Chinese) is a free, open-source native macOS app published on GitHub by the account andyhuo520, a single developer with no company named on the project. It manages the Anker soundcore AI recorder "bean" (model D3200): it pulls recordings to the Mac, transcribes them, turns them into notes and to-dos, and lets you ask questions across everything you have recorded. The repository was created and its first release, v0.1.0, was published on 16 November 2026; the commits are signed Berryxia.
Transcription runs on the Mac with Qwen3-ASR 0.6B (from the Qwen family) through llama.cpp, with macOS speech recognition as the fallback, and the README says recordings are not uploaded. Notes can be written by a local Qwen3-4B model, a local Ollama or LM Studio server, or a cloud service you give your own key. The app also ships a command-line tool and an MCP server, so Claude Code, Codex and OpenClaw can search recordings, read notes and to-dos, and save text or audio into the library.
It fits people who record on a Mac and want the archive to stay local: consultants logging client calls, researchers and journalists transcribing interviews, and language learners who replay podcasts one sentence at a time with an English study mode that grades vocabulary. It also works without the recorder, using the Mac microphone, system audio, imported files or links from YouTube, Bilibili and Apple Podcasts. Notes export as Markdown with frontmatter, so they drop into Obsidian.
Some buyers should look elsewhere. BeanPod needs macOS 26, and macOS warns about an unidentified developer when you first open it. The author tested it on one recorder firmware only and calls it a personal project for learning, research and testing, so there is no support promise. Plaud sells a recorder and a supported app with cloud sync, while Otter and Fireflies are hosted meeting note takers that work across devices. More recorders and note takers are listed under AI note-taking apps.
The v0.1.0 download is a single DMG for Apple silicon and Intel Macs. The interface is in Simplified Chinese and English, the iPhone version in the repository is still experimental, and support for the DeDao AI recording card follows a published protocol and is untested on hardware.
Our call: choose BeanPod if you own the recorder or want free on-device transcription and are comfortable running an unsigned 0.1 app. Choose Plaud or Otter if you want a service that someone supports.
Product facts on this page (pricing, plans, integrations, compliance) are read from andyhuo520's own pages and documentation. The verdict is Hokai's reading of them, not the vendor's claim.