Krisp is two products in one app: noise cancellation on your microphone and on what you hear, and a note taker that records calls without joining them. The note taker is within reach. Capture the microphone and the call audio, transcribe on the machine with Whisper, and ask an LLM for the summary; Meetily and anarlog already do this in the open. The weekend goes on audio plumbing: a loopback device on macOS or Windows, the permission prompts, and the call apps that behave differently. The noise cancellation is the part that does not transfer, because Krisp's trained models and virtual devices that work in every call app are the product.
Before you buildKrisp's 7-day trial covers every feature, so try noise cancellation there before deciding what to build. The DIY version records and summarizes but does not clean audio.
Build me a local meeting recorder and note taker to replace Krisp. Requirements: - Python 3.12: a FastAPI server with one plain HTML page and no build step. `python -m meetnotes` serves http://localhost:4900; Start and Stop buttons begin and end a meeting. - Capture: two input devices read with `sounddevice`, the microphone and a loopback device (BlackHole 2ch on macOS, a WASAPI loopback on Windows), named in .env as MIC_DEVICE and LOOPBACK_DEVICE. Save each as 16 kHz mono WAV in ~/MeetingNotes/<date>-<title>/. - Transcription: `faster-whisper` with WHISPER_MODEL from .env (default `small`, CPU int8), run on 30-second chunks while recording. Label lines "You" for the mic and "Them" for the loopback, with timestamps. - Notes: after Stop, send the transcript to an OpenAI-compatible endpoint from .env (Ollama on localhost by default) and write notes.md with a summary, decisions and action items; keep the raw text in transcript.md. - Storage: sqlite3 with FTS5 in ~/MeetingNotes/index.db (title, start, duration, folder path) and a search box over transcripts and notes. - Interface: a meeting list, a meeting page with transcript, notes and an audio player, export as .txt or .md, and delete a meeting together with its files. - Single user. No accounts, no telemetry, local-first: only the LLM endpoint you configure leaves the machine, and the README says exactly what is sent. - Out of scope: noise cancellation and a virtual microphone, accent conversion, calendar auto-join, speaker separation beyond You/Them, CRM integrations. - README: install steps, loopback setup (budget half an hour for it), the .env keys, and a note to tell people before recording and to check local recording-consent rules.
Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.
The first prompt builds the core. Run these one at a time in the same folder to grow it into your own Krisp.
Add an importer to the meeting notes app in this folder. Requirements: - A command `meetnotes import-krisp <folder>` that reads the .txt transcripts and .mp3 recordings downloaded from Krisp, pairing files by name. - Create one meeting per pair in ~/MeetingNotes with the original date taken from the file's modified time, index it in FTS5, and skip files that were imported before. - Offer to generate notes.md for each imported transcript with the configured LLM, and print a count of imported, skipped and failed files. - Update the README with the Krisp export steps.
Add a question box to the meeting notes app in this folder. Requirements: - Retrieve the 8 best transcript chunks for a question with FTS5 bm25, then send only those chunks to the configured LLM and show the answer with links to the meetings it used. - Never send a whole transcript; show the user the chunks that will be sent before the call. - Keep a history of questions in SQLite and let the user delete any entry. - Update the README with exactly what leaves the machine.
Add a weekly digest to the meeting notes app in this folder. Requirements: - A command `meetnotes digest` that collects action items from notes.md files of the last 7 days and writes ~/MeetingNotes/digests/<year>-W<week>.md grouped by meeting. - Mark items found twice across meetings as repeated. - Optionally email the digest through SMTP settings in .env. - Update the README with a cron example.
People pay for clean audio that works in whatever call app they are in, with no bot in the meeting and notes waiting after the call. The security paperwork (SOC 2, HIPAA, a promise not to train on meeting data) matters when a company has to approve a recorder.
Estimate · Whisper and Ollama run on your own machine for nothing; a hosted LLM for summaries costs a few cents a meeting. The build is token usage on your agent plan.
| Plan | Monthly | Annual, per month | What you get |
|---|---|---|---|
| Free trial | $0 | $0 | 7 days of every premium feature: transcription, noise cancellation, recording and AI notes, with accent conversion limited. |
| Core | $16 | $8 | AI note taking including in-person meetings, transcription, recording, noise cancellation, multilingual transcripts, AI chat, mobile app, outbound call recording, integrations and webhooks, 1 hour a day of accent conversion, 10 GB storage. |
| Advanced | $30 | $15 | Everything in Core plus more accent conversion, Salesforce and ConnectWise, admin controls, manager view, deal grouping, trim and comment, team webhooks, 60 GB storage. |
| Enterprise | custom | — | SSO and SCIM, super admin, on-device private transcription and recordings, HIPAA support, usage analytics, unlimited storage, dedicated account manager. |
The pricing page offers only a 7-day trial with no card; we could not confirm a permanent free plan.
Kinda. The core is a weekend build with the prompt on this page, but real gaps remain: Noise cancellation on the microphone and the speaker, which is what Krisp started with, Accent conversion and voice translation.
About $16 a month on the Core plan per seat, $192 a year (checked 2026-10-19).
Noise cancellation on the microphone and the speaker, which is what Krisp started with; Accent conversion and voice translation; Automatic notes for meetings on your calendar, plus the mobile app; Hosted transcripts that teammates can open, comment on and share; Connectors to Slack, Notion, HubSpot and Salesforce, and a managed security story for a team. If any of those matter to you, keep paying.
Yes: Meetily (open source), anarlog (open source), Buzz (open source).