Which platforms does the bot support?

Zoom, Google Meet and Microsoft Teams. For anything else, such as an in-person meeting or a call on another tool, upload the recording instead: audio or video up to 2 GB, with the same write-up as a live call.

How long does a write-up take?

Usually under two minutes after the call ends, for a meeting of ordinary length. Longer meetings take longer because there is more to read.

Can I fix a wrong action item?

Yes. Every action item can be edited, reassigned, given a different date or deleted. The suggested owner is a suggestion, and confirming it is a human decision.

What happens if processing fails?

The meeting is marked failed with a reason and a retry button. Automatic retries run in the background with backoff, and give up after three attempts rather than looping forever.

Do you train models on our meetings?

No, and our processing providers are contractually barred from doing so. Nobody at Notepik reads your transcripts.

Does everyone in the workspace see every meeting?

Yes, within a workspace. Notepik is built around a shared library rather than private notebooks. If you need separation, use more than one workspace: a meeting is never visible outside the one it was recorded in.

Can I stop recording partway through a call?

Yes. Remove the bot at any point, and the partial recording can be deleted from the meeting page. If somebody objects mid-call, that is the right thing to do rather than deleting it afterwards.

What happens to a recording when I delete a meeting?

The recording, transcript, summary and action items go immediately, including the media file in storage. It is not queued for later.

Features

Everything between the call ending and the work getting done

Notepik is not a recorder with a summary bolted on. It is built around the two things a meeting actually produces: what was decided, and who agreed to do what.

At a glance

Bot or upload

live calls or existing files

Speaker labels

who said what, when

Summary and decisions

separate, not one blob

Action items

owner and due date

Cross-meeting search

every line, not just titles

Comments and mentions

anchored to a moment

Get the meeting in without anyone installing anything

Paste the link and the bot joins, or upload a recording you already have. Either way the write-up is waiting when you get back to your desk.

A bot that joins the call: Paste a Zoom, Google Meet or Microsoft Teams link and the bot joins on time, under a name you choose, in the participant list. Recording people who do not know is not a feature we build.
Or upload a recording you already have: An in-person meeting recorded on a phone, a webinar, a call on a platform the bot does not join. Upload the audio or video, up to 2 GB, and it gets the same speaker labels, summary and action items as a live call.
Speaker-labelled transcription: Every line is attributed and timestamped, so a summary claim can be traced back to the moment somebody actually said it.
Whatever language the meeting was in: Transcription is not English-only. The write-up comes back in the language the model was asked for.

Understand

Three outputs, because a wall of text is not notes

The summary is for the person who missed it. The decisions are the record. The action items are the work. Keeping them separate is what makes the page scannable in fifteen seconds.

Summary, decisions and action items: Three separate outputs, not one blob. An executive summary for people who missed it, the decisions as a list, and the follow-ups with a suggested owner and a due date.
The right model for the meeting: A fifteen minute standup does not need the same model as a two hour strategy session. Short meetings use a fast, cheap tier; long or crowded ones are escalated automatically.
Traceable back to the transcript: Action items carry the segment they came from. When somebody disputes a follow-up, you can read the ten seconds that produced it rather than arguing from memory.
Cost tracked per meeting: Every write-up records which model ran and what it cost, so usage against your plan is a number rather than an estimate.

Collaborate

A shared library, not one person's notes

Everything lands in a workspace the whole team can read, search and argue with. Meeting output that lives in one person's document is lost the moment they get busy.

Search what was said, not just titles: Full-text search runs across meeting titles, summaries and every transcript line, with the matching phrase highlighted and a timestamp to jump to.
Comments anchored to a moment: Reply to a specific line of the transcript rather than to the meeting in general. @mention a colleague and they get an email with the comment in it.
Tags across the library: Group meetings by client, project or theme, and filter the whole workspace by one.
Everything is shared by default: Output lands in the workspace, not in the notebook of whoever set the recording going. Nothing is lost when one person gets busy.

Reliability

The unglamorous part that decides whether you keep using it

Most of what can go wrong here is somebody else's outage: a missing webhook, a model timing out, Slack having a bad afternoon. None of those should cost you a meeting, so each one has something behind it that notices and recovers.

Nothing waits on a webhook: Every stage has a poller behind it. If a callback from the recording provider goes missing, a job reaches the same conclusion by asking instead, within about two minutes.
Failed write-ups retry themselves: With backoff, three times, then it stops and puts a retry button in front of a human. The transcript is never re-paid for: a retry costs one model call.
Integration pushes retry too: If Slack is down for the ten seconds your meeting finished, the push is queued and tried again with growing gaps, and abandoned after six attempts rather than hammering a revoked token forever.
A failure never loses your meeting: The transcript, summary and action items are saved before anything is pushed or emailed. A broken integration cannot turn a successful meeting into a failed one.

Notifications

Enough email to be useful, and no more

Deliberately little. Anything a scheduled job can send is de-duplicated before it goes out, so a daily job cannot mail you the same reminder three days running.

The meeting is ready: Goes to everyone in the workspace with the summary and the decisions, once. A retried pipeline does not email the whole team again.
Something is yours: Assigning an action item emails the person it landed on, and mentioning somebody in a comment emails them. Never the person who did the assigning.
Daily or weekly digest: Opt-in, and set per workspace rather than per person, so being in three teams does not mean three times the email. A digest with nothing in it is never sent.
Before you hit a limit: The owner is emailed at 80 percent of the month's hours, and three days before a trial ends. Once, not every day.

Administration

Roles, records and a clean way out

The parts that start mattering once more than three people share a workspace and somebody has to answer for what changed.

Four roles, narrowly scoped: Owner, admin, member and viewer. Billing is owner-only, integration credentials are admin-only, and a viewer reads without being able to change anything.
An audit trail for admin actions: Role changes, suspensions, integration connections and plan edits are recorded with who did them and what changed, including a before and after where it matters.
Seats enforced before they are used: A pending invite counts against the limit, so a ten seat plan cannot quietly become twelve the moment everybody accepts.
Leave without losing anything by accident: Closing a workspace locks it immediately and cancels the subscription, but keeps everything for 30 days so the owner can reopen it. After that it is deleted for good.

Trust

Your meetings stay yours

Workspace isolation is enforced by Postgres row-level security rather than by application code remembering to filter, third-party credentials are encrypted separately, and nothing is ever used to train a model.

Isolation the database enforces: A query that forgets to filter returns nothing rather than somebody else's meeting. We test this rather than assume it.
Credentials encrypted at rest: The Slack, Asana, Trello and ClickUp tokens a workspace connects are encrypted with a key held outside the database.
Recording people know about: The bot is always visible in the participant list. There is no hidden mode, deliberately.
Never training data: We do not train models on your meetings and our providers are contractually barred from it. Nobody here reads your transcripts.

The short answers

Capture: Bot for Zoom, Google Meet and Teams, or upload a file
Upload formats: mp3, m4a, wav, flac, mp4, mov, webm, up to 2 GB
Transcription: Speaker-labelled and timestamped
Languages: Not English-only
Typical turnaround: Under two minutes after the call
Search: Titles, summaries and every transcript line
Integrations: Slack, Asana, Trello, ClickUp
Roles: Owner, admin, member, viewer
Authentication: Email and password, or Google

Hours and seats depend on your plan, and those are on the pricing page.

What Notepik does not do

Worth ten seconds now rather than a trial and two emails. If one of these is a dealbreaker, we would rather you found out here.

No playback in the app

The meeting page shows the transcript rather than a player. Notepik is transcript-first: the searchable text is more useful than scrubbing an hour of video.

No SAML single sign-on or SCIM

No SOC 2 and no HIPAA agreement

No certification and no third-party penetration test yet. We do not sign business associate agreements, so do not put protected health information through it.

Where we would actively advise against using it, and why, is on the use cases page.

Questions about how it works

Which platforms does the bot support?: Zoom, Google Meet and Microsoft Teams. For anything else, such as an in-person meeting or a call on another tool, upload the recording instead: audio or video up to 2 GB, with the same write-up as a live call.
How long does a write-up take?: Usually under two minutes after the call ends, for a meeting of ordinary length. Longer meetings take longer because there is more to read.
Can I fix a wrong action item?: Yes. Every action item can be edited, reassigned, given a different date or deleted. The suggested owner is a suggestion, and confirming it is a human decision.
What happens if processing fails?: The meeting is marked failed with a reason and a retry button. Automatic retries run in the background with backoff, and give up after three attempts rather than looping forever.
Do you train models on our meetings?: No, and our processing providers are contractually barred from doing so. Nobody at Notepik reads your transcripts.
Does everyone in the workspace see every meeting?: Yes, within a workspace. Notepik is built around a shared library rather than private notebooks. If you need separation, use more than one workspace: a meeting is never visible outside the one it was recorded in.
Can I stop recording partway through a call?: Yes. Remove the bot at any point, and the partial recording can be deleted from the meeting page. If somebody objects mid-call, that is the right thing to do rather than deleting it afterwards.
What happens to a recording when I delete a meeting?: The recording, transcript, summary and action items go immediately, including the media file in storage. It is not queued for later.

The only test that matters is your own meetings

Summaries read well on someone else's demo call. Run it on one of yours and see whether it caught the thing you would have written down.

Start free Read the guide See it by team