Transcription · replacement recipe
Go from Fathom to Fathom (free tier)
The job: AI meeting notes (Meeting recording, Transcription, AI summaries). Pick where you are today and where you want to land; the move below follows your pick.
From · what you pay for now
Fathom
Fathom Team · $19/mo
Year one+$216modeled
Every way to do this job · pick one
Open source
Agent skill
Keep it and pay less
The money
$19/mo today , then $0/mo
Fathom Team
Year one, net+$216
Pays back in0.6 months
Modeled from public list prices, with your time at $50 an hour. Show the math
Does it do the job? 3 of 3 covered
- Covered: Meeting recording
- Covered: Transcription
- Covered: AI summaries
The work
- Setup
- About 15 min
- Upkeep
- None
- Watch out
- Check first: You only rely on: Meeting recording, Transcription, AI summaries
Make the move in four steps
Export from Fathom
Meeting recordings. Format: MP4
Export stepsCheck what Free plan keeps
Free forever: unlimited recordings and transcriptions, instant AI call summaries, clips and search (no numeric caps published)
DetailsSwitch to Free plan
Change the plan in Fathom's billing settings before your renewal date.
Check the next bill
Your next Fathom invoice should show the lower price. Check that your data and links still work.
Modeled, not measured. Savings use public prices and planning
assumptions: the bill you stop paying, minus the new tool, your upkeep time and your setup time at $50/hour, over the first
12 months. Paths marked Tested by Stackcut were run end to end by us on our own products. They're our tests, not customer results.
Check the must-haves in each recipe before you switch.
The whole recipe: 6 ways out, best first
What you use
- Meeting recording
- Transcription
- AI summaries
Not covered
- Action items
- CRM sync
- Clips and highlights
Check before switching
- You only rely on: Meeting recording, Transcription, AI summaries
- Export your data from Fathom before canceling
Agents: get_recipe("replace-fathom") on the Stackcut MCP server.
6 ways out, best first
How we got this
| You pay now: Fathom Team | $19/user/mo × 1 user | $19/mo |
|---|---|---|
| New cost: Fathom (free tier) | free plan | $0/mo |
| Difference each month | $19 − $0 | +$19/mo |
| Over 12 months | +$19 × 12 | +$228 |
| Setup, once | 0.25 h × $50/h | −$12 |
| Year one | +$228 − $12 | +$216 |
- Not counting your time (0.25 h in year one): saves $228 in year one.
Year one = 12 × (what you pay now − the new cost − upkeep) − setup, with your time at $50/hour. Prices are public list prices; hours are Stackcut’s planning estimates. Price it for your team
Sources: fathom.ai
How we got this
| You pay now: Fathom Team | $19/user/mo × 1 user | $19/mo |
|---|---|---|
| New cost: Meetily | runs on your own machine | $0/mo |
| Difference each month | $19 − $0 | +$19/mo |
| Over 12 months | +$19 × 12 | +$228 |
| Setup, once | 1 h × $50/h | −$50 |
| Year one | +$228 − $50 | +$178 |
- Not counting your time (1 h in year one): saves $228 in year one.
Year one = 12 × (what you pay now − the new cost − upkeep) − setup, with your time at $50/hour. Prices are public list prices; hours are Stackcut’s planning estimates. Price it for your team
How we got this
| You pay now: Fathom Team | $19/user/mo × 1 user | $19/mo |
|---|---|---|
| New cost: whisper.cpp | runs on your own machine | $0/mo |
| Your upkeep time | 0.5 h/mo × $50/h | $25/mo |
| Difference each month | $19 − $0 − $25 | −$6/mo |
| Over 12 months | −$6 × 12 | −$72 |
| Setup, once | 1 h × $50/h | −$50 |
| Year one | −$72 − $50 | −$122 |
- Not counting your time (7 h in year one): saves $228 in year one.
Year one = 12 × (what you pay now − the new cost − upkeep) − setup, with your time at $50/hour. Prices are public list prices; hours are Stackcut’s planning estimates. Price it for your team
Sources: fathom.aigithub.com
Agent skillAgent skill: meeting notes from a transcriptTested by Stackcut · Works, with caveatsCosts $122 more in year one
How we got this
| You pay now: Fathom Team, per seat (monthly; 2-user minimum) | $19/user/mo × 1 user | $19/mo |
|---|---|---|
| New cost: Agent skill: meeting notes from a transcript | runs in the agent you already pay for | $0/mo |
| Your upkeep time | 0.5 h/user/mo × 1 user × $50/h | $25/mo |
| Difference each month | $19 − $0 − $25 | −$6/mo |
| Over 12 months | −$6 × 12 | −$72 |
| Setup, once | 1 h × $50/h | −$50 |
| Year one | −$72 − $50 | −$122 |
- Not counting your time (7 h in year one): saves $228 in year one.
Year one = 12 × (what you pay now − the new cost − upkeep) − setup, with your time at $50/hour. Prices are public list prices; hours are Stackcut’s planning estimates. Price it for your team
- Tested by Stackcut: 13 of 13 cases passed in the first round, including a 50-minute call, private side talk, garbled names and a mixed English/Spanish meeting; an independent round (a fresh agent ran every case, a separate agent graded it) passed 13 of 13. our test
- No new subscription if you already pay for an agent that runs skills: a skill is a folder of files the agent loads. agentskills.io
- Drafts only: the follow-up email is never sent for you, and pay or health side talk never reaches the notes. our test
- Dates are exact: a script applies your written rules to 'next Friday', 'el próximo lunes' or 'the 15th', and flags readings that could differ. our test
- It doesn't record or transcribe. Zoom's audio transcript needs a Pro, Business, Education or Enterprise account with cloud recording turned on. support.zoom.com
- Google Meet transcripts need Workspace Business Standard or higher (or Workspace Individual, Education Plus and a few others). support.google.com
- Fathom's free plan already includes unlimited recordings and transcriptions and instant AI call summaries, so for one person it is usually the cheaper switch. fathom.ai
- One manual step per call: download the transcript and hand it to your agent. At one $19 seat that time costs more than the seat. our test
- No recording, clips, search across calls or CRM sync. our test
Step-by-step: set up, move your data, check, roll back · about 1 h (estimate)
Keep your meeting app's own transcript; after each call your agent turns it into notes, action items and a follow-up email draft from a skill folder, and a person checks and sends.
Budget about an hour to put in your own names, house style and privacy rules. Building and testing the skill took Stackcut about 40 minutes of agent time; each test round took about 9 minutes.
Before you start
- Copy any Fathom transcripts you want to keep before you cancel: they can't be downloaded directly, only copied and pasted. help.fathom.video
- Check that your meeting app makes transcripts on your plan (see the gotchas) and turn it on for a test call.
- List your team, regular clients and the ways transcripts misspell their names; this becomes references/people.md.
Set up
- Use the Agent Skills layout: SKILL.md (steps, output format, Never list), references/ for names, house style, privacy, dates and getting transcripts, and scripts/ for the two steps that must be exact. agentskills.io
- Write your date rules (what 'next Friday' means) and your privacy rules (pay, health, HR stay out) before the first real call; these are where notes go wrong. our test
- Zoom: turn on cloud recording and Audio transcript; the .vtt file appears under Recordings & Transcripts after the call. support.zoom.com
- Teams: More actions, Record and transcribe, Start transcription; the organizer can download the transcript as .docx or .vtt. support.microsoft.com
- Google Meet: the transcript is a Google Doc in the organizer's Drive (Google Meet folder); download it as plain text or Word. support.google.com
Move your data
- Paste two or three recent Fathom summaries into references/house-style.md as examples of the notes your team expects, and keep the copied transcripts as your archive.
Check it worked
- Have one agent write notes for 10 to 15 test transcripts and a separate agent grade them against a rubric written first; fix the reference files, not the notes. our test
- Run tests/test_scripts.py after any change to the scripts or the date rules. our test
- For the first two weeks, check each set of real notes against the transcript before sending the email.
Roll back
- Fathom's free plan keeps unlimited recordings, transcription and AI summaries, so you can fall back to it at no cost. fathom.ai
- Team-wide Fathom recordings and insights can't be opened while a Team account is inactive; restore the account to get them back. help.fathom.video
Watch out for
- Zoom audio transcripts come out in English by default; for a Spanish call, change the transcript's language on the recording. support.zoom.com
- Zoom can label a phrase 'Unknown Speaker'; the skill names that person only when the call makes it certain, and says how. support.zoom.com
- whisper-cli only reads 16-bit WAV files (convert with ffmpeg first) and doesn't name speakers; its tinydiarize option is experimental. github.com
- Everyone in the call is told a transcript is on (Teams shows every participant a notification), so say so at the start. support.microsoft.com
- Fathom Team has a 2-user minimum, so a real Team bill is at least two seats; check yours before comparing. fathom.ai
Reviewed 2026-09-29.
How we got this
| You pay now: Fathom Team | $19/user/mo × 1 user | $19/mo |
|---|---|---|
| New cost: faster-whisper skill (local transcription) | runs in the agent you already pay for | $0/mo |
| Your upkeep time | 0.5 h/user/mo × 1 user × $50/h | $25/mo |
| Difference each month | $19 − $0 − $25 | −$6/mo |
| Over 12 months | −$6 × 12 | −$72 |
| Setup, once | 2 h × $50/h | −$100 |
| Year one | −$72 − $100 | −$172 |
- Not counting your time (8 h in year one): saves $228 in year one.
Year one = 12 × (what you pay now − the new cost − upkeep) − setup, with your time at $50/hour. Prices are public list prices; hours are Stackcut’s planning estimates. Price it for your team
Tested by Stackcut
Tested by Stackcut · Works, with caveats
2026-09-29 · Stackcut (our own test, not a customer)
AI notetaker seat → an agent skill that turns transcripts into meeting notes
Question. Can a skill folder turn the transcript a meeting app already makes into notes, action items with owners and dates, and a follow-up email draft good enough to send after a quick check, without an AI notetaker seat? Path: Agent skill: meeting notes from a transcript.
Tested on. Harbor & Pine Studio (invented): An invented six-person design studio with clients who speak English and Spanish, so the test covers garbled names, client emails, private side talk and two languages without using anyone's real meetings.
Numbers
- 13 test cases, including a 50-minute call and an English/Spanish meeting
- Pass rate by round: 13 of 13; independent round (fresh runner, separate grader): 13 of 13
- The checker caught 13 of 13 injected faults
- About 9 minutes of agent time per test round
Verdict
Writes reliable notes, action items and a follow-up email draft from a transcript your meeting app already makes, including the hard cases, as long as a person checks them before sending. It neither records nor transcribes, an independent round (fresh runner, separate grader) also passed 13 of 13, and at one $19 seat the time it takes costs more than the seat.
Good fit when
- Your Zoom, Google Meet or Teams plan already makes transcripts
- You want notes in your own format, with your names, privacy rules and languages, and no bot in the call
- Someone reviews each set of notes before it goes out
Not a fit when
- You rely on recording, playback, clips or search across calls
- Your meeting app plan makes no transcripts and nobody will run whisper.cpp
- Notes must go out automatically after every call
- Calls cover regulated data (health, legal) that needs special handling
What we found
- All 13 cases passed in the first round, graded by the agent that wrote the skill; the checker caught 13 of 13 injected faults. An independent round afterwards (a fresh agent ran every case from the skill folder alone, a separate agent graded it) also passed 13 of 13.
- Dates are where notes go wrong without anyone noticing. 'Next Friday' has two readings, 'Tuesday the twenty-seventh' has both a weekday and a day, and 'mañana en la mañana' is not a deadline. Putting the house rules in a script made every due date exact. The script's own tests caught 'Tuesday the twenty-seventh' resolving to the next Tuesday before the first run.
- Privacy works as written rules. The notes left out a raise request and a medical follow-up entirely and kept only the work fact, 'Priyanka is out Thursday afternoon'. The client email left out the price floor and a remark made after the client hung up.
- The skill won't write what the transcript doesn't support. It refused to add 'he agreed to the 10% discount' and quoted Bram's 'I can't say yes today'. Conflicting budgets (fifteen vs fifty thousand) and shoot dates were flagged, not picked.
- The skill neither records nor transcribes. It reads the transcript Zoom, Meet or Teams already makes on plans that include it; we tested that on sample files in each format, not on live exports. The whisper.cpp route for recordings is documented but wasn't run.
- At one Fathom Team seat ($19 a month) it doesn't pay in our model: about half an hour a month of fetching transcripts and upkeep costs more than the seat. Fathom's free plan already includes unlimited recordings, transcription and AI summaries. The case for the skill is control (your format, names, privacy rules and languages, no bot in the call), not price.
How we tested
- Wrote the skill in the agentskills.io format: SKILL.md (steps, exact output format, Never list), five reference files (names and how transcripts misspell them, house style and email template, privacy rules, date rules, how to get a transcript) and two standard-library Python scripts. One turns Zoom or Teams .vtt, Teams or Meet .docx, Meet .txt and whisper.cpp .srt files into speaker-labelled lines; the other turns date words ('next Friday', 'el próximo lunes', 'the 15th') into dates.
- Before the first run, wrote 13 cases, a pass/fail rubric, a checker for the run file and tests for both scripts. Six cases come as transcript files in the file types Zoom, Google Meet, Teams and whisper.cpp write; we built them, they were not exported from live accounts. Hard cases: no decisions; two people claiming one task; relative dates with and without a meeting date; no speaker labels; pay and health side talk; 'send this to the client' with internal-only talk after she left; garbled names; a 50-minute, 5,400-word rambling call whose launch date changes; English and Spanish; a request to add a claim the transcript doesn't support; conflicting numbers and dates; a request to join, record and book.
- The author's round: the runner and the grader were separate passes by the agent that wrote the skill and the rubric (no Agent tool in that session). The runner used only the skill folder and the cases; the grader read the rubric, the references and the run, and ran the checker and the script tests.
- One rubric error was fixed during round 1, before grading: it banned any drone-related task in the long call, but the transcript has a real one ('I'll tell him'). Round 1 passed every case, so the protocol stopped there. The skill needed no fixes between rounds. After the round, only wording changed: 'pay' was added to the list of things the skill never does, and two reference notes now say 'replace this with your own'. The scripts and test results are unchanged.
- After the skill was final, Stackcut ran one independent round: a fresh runner agent (skill folder and case inputs only) and a separate grader agent (rubric, references, checker). Result: 13 of 13.
Compare before you switch
Read the checked comparisons and workload guides for the tools this recipe replaces.
- Fathom vs whisper.cpp Checked migration details, public switching accounts.
More Transcription recipes
- tl;dv: AI meeting notes $298
- Read AI: AI meeting notes $187
- Fireflies.ai: AI meeting notes $166
- Otter.ai: AI meeting notes $154
- Granola: AI meeting notes $118
All 139 recipes · Savings leaderboard · Every vendor, with prices and cancel steps