Zoom meeting notes in Obsidian without a bot or recording
By Kyle Nelson, Founder, Fazit
If you want Zoom meeting notes in Obsidian, the cleanest setup is a Mac app that writes Markdown straight into your vault instead of syncing from a meeting-note database. Fazit does that for 1:1 Zoom client calls: no bot joins, transcription and note generation run on your Mac, and the call audio is never written to disk.
That last part decides the workflow for consultants, coaches, lawyers, financial advisers and therapists. A pretty note is easy to copy. A note pipeline that never creates an audio file is the difference between keeping one Markdown artifact and creating a second artifact you now have to protect.
How do I take Zoom meeting notes in Obsidian?
Use a workflow that ends in a plain .md file inside your Obsidian vault. That sounds obvious, but many meeting-note tools treat Obsidian as an export target after the real note has already been created and stored somewhere else.
Fazit takes the opposite route. You set an Obsidian vault folder in the macOS menubar app, start capture for the Zoom call, stop when the call ends, and Fazit writes one Markdown note into the vault. The source statement describes the output as a plain Markdown file written into a folder of your Obsidian vault, with no Fazit database holding a copy.
That matters because Obsidian is file-based. Your note belongs in a folder you control, not as a row in a vendor database that later syncs into your vault. We wrote more about that file-first model in Obsidian meeting notes should be files, not rows in a vendor's database, and the same rule applies to Zoom calls.
For a 1:1 client call, the finished note can include a summary, moments from the call, action items with owners, a You/Them transcript, and a paste-ready follow-up email.
Can Zoom meeting notes be saved as Markdown files?
Yes. The question is whether Markdown is the original output or an export copy.
If the meeting-note tool stores your note in its cloud database first, Markdown is just a delivery format. You still have to ask where the transcript went, where the audio went, what database holds the note, what subprocessors saw the content, and what deletion actually removes.
Fazit writes the Markdown file directly into your Obsidian vault or Apple Notes. There is no separate Fazit database holding a copy of the note. For Obsidian users, that means the vault is the system of record, and the file can be searched, linked, renamed, moved, backed up and versioned like any other note.
Granola users often solve the same problem with community sync plugins, cache scrapers or copy-paste routines, because Granola has no official Obsidian export according to the source material for How to sync Granola notes to Obsidian. That works until sync becomes the workflow. Direct Markdown output removes the sync layer entirely.
What should a Zoom to Obsidian workflow avoid?
Avoid four things: a visible bot, cloud transcription, an audio file, and a vendor database as the original home of the note.
A bot changes the social shape of a client call. It joins as another participant, and everyone sees it. If your client asks why a recording assistant joined the room, the meeting has already become about the tool.
Cloud transcription changes the data shape. The call audio leaves your machine, reaches a vendor or a transcription provider, and becomes part of someone else's infrastructure. That may be acceptable for some internal meetings. It is harder to justify for a lawyer, therapist, adviser, coach or consultant speaking with a client under a confidentiality obligation.
An audio file changes the evidence shape. Once a recording exists, it can leak, be retained, be deleted incompletely, or be requested later. The cleaner artifact is the note you meant to keep, without a recording behind it.
A vendor database changes ownership in practice. You may have a copy in Obsidian, but the original system still lives elsewhere. For confidential work, the vault should hold the note, not mirror it.
How does Fazit make Zoom notes without a bot?
Fazit captures at the audio layer on the Mac. A Core Audio process tap reads the call app's output and the microphone directly, so nothing joins the Zoom meeting and no other participant sees anything.
That design also avoids calendar access. A meeting bot usually needs to know what to join, and the path to that is your calendar. Fazit does not need that grant for capture because you pick the call app locally. The same mechanism works with Zoom, WhatsApp Desktop, Meet in a browser and Slack via the app picker, according to the repository README.
The architecture is narrower than a general meeting assistant, on purpose. Fazit is built for 1:1 client calls. Your microphone is one stream, labeled "You." The call app's audio is the other stream, labeled "Them." Because those streams are physically separate, attribution in a 1:1 call does not depend on speaker diarization guessing who talked.
That constraint is honest. In group calls, remote speakers are mixed into one system-audio channel, and separating them requires diarization. The research document reports open local diarization error rates of 13.3% to 17.7%, which is the reason the MVP cuts multi-party speaker labels.
Does Fazit record Zoom calls?
No. Fazit's stated architecture is that audio is never written to disk. It exists only in a fixed-size RAM ring buffer with no write API, and the buffer is destroyed on every exit path including errors.
The source statement is specific: every note records audio_retained: false in its frontmatter. Transcription runs on-device using Parakeet via CoreML, and note generation runs through a local model on localhost. Neither the audio nor the transcript leaves the machine for note generation.
That does not mean every network request is absent. Fazit's own accuracy notes separate call content from account infrastructure. Models are downloaded once from public CDNs on first run, and account, licensing and payment use Fazit servers. Account, licensing, updates and optional product analytics are separate from the call-content path.
We have a longer architecture piece on the same distinction in on-device vs. cloud AI notetakers. For Zoom meeting notes in Obsidian, the short version is simpler: the note is kept, the audio file is never created.
Zoom meeting notes Obsidian comparison
The right comparison is not "which tool writes the nicest summary?" Good summaries are now common. The better comparison is where the Zoom call content lives after the call.
| Workflow | Bot joins Zoom | Where the note starts | Audio or transcript path | Obsidian result |
|---|---|---|---|---|
| Manual notes in Obsidian | No | Your vault | No tool captures audio | Native Markdown, but you write everything yourself |
| Zoom AI Companion | No meeting bot described in the source material | Host's Zoom account | The source material says it creates a transcript, summarizes it, and stores it on the host's account | Not an Obsidian-native Markdown workflow |
| Granola plus community sync | No bot | Granola first | The source material says Granola streams audio to third-party transcription services | Sync, scraping or copy-paste into Obsidian |
| Fazit | No | Your Obsidian vault | Audio stays in RAM, transcription and note generation run on the Mac, audio is never written to disk | Native Markdown file in your vault |
The table has a bias because the problem has a bias. If your job involves confidential 1:1 calls, the safest note workflow is the one with the fewest durable artifacts.
Is this legal for client calls?
This is not legal advice, and consent and professional-conduct duties apply to the person running the call.
Architecture can reduce what the vendor receives and what artifacts exist, but it does not remove your duty to tell the other person what you are doing where consent rules or professional duties require that. Fazit's repository says the architecture reduces the vendor's and the audio-file exposure, but does not remove the user's consent duty.
The product repository gives two examples: California CIPA is all-party, and Germany's §201 StGB is criminal. The same repository also says users should say a plain consent line before a call, because the other person should have a real chance to say no.
For profession-specific context, we have separate pieces for lawyers, financial advisors and consultants. The practical rule is the same across them: keep the note you need, and do not create a recording unless you have decided you need and can justify a recording.
What should be in an Obsidian client-call note?
A client-call note should be useful without replaying a recording. If you need the audio later to understand the note, the note failed.
Fazit's repository describes the finished output as a session note with a summary, moments from the call, action items with owners, a paste-ready follow-up email, and a You/Them transcript. That is the right set for recurring 1:1 work because it maps to what happens after the call: update the client record, remember the commitments, send the follow-up, and prepare for the next session.
The You/Them transcript is especially useful in Obsidian because it becomes local memory. You can link it to a client page, search for commitments, and keep the note beside your other work. The artifact is a Markdown file in a vault, not a playback page in a vendor account.
Fazit requires macOS. The site statement says there is no Windows, Linux, iOS or Android version. It also makes no maturity, scale or certification claims, and holds no SOC 2 or HIPAA certification.
What is the best setup for Zoom meeting notes in Obsidian?
For confidential 1:1 client calls, the best setup is direct Markdown output plus local processing. Use a tool that captures Zoom locally, transcribes on the Mac, writes one .md file into your vault, and never writes call audio to disk.
That is the narrow job Fazit is built to do. It is a native macOS menubar app for Zoom and other desktop call apps, with bot-free capture, on-device transcription, local note generation and Obsidian output. It does not try to be a CRM, a meeting archive, a team database or a video library.
The product choice is opinionated: your Obsidian vault should contain the finished note, and the call audio should not exist as a file anywhere.
Sources
- Fazit site statement and accuracy notes: https://getfazit.com/llms.txt
- Fazit security and architecture page: https://getfazit.com/security/
- Core Audio process-tap sample, AudioCap: https://github.com/insidegui/AudioCap
- FluidAudio Swift SDK and Parakeet CoreML reference: https://cocoapods.org/pods/FluidAudio
- Apple developer documentation referenced for Core Audio taps: https://developer.apple.com/