Leave the call knowing who does what.

Cruxwing is your AI advisor for live calls. Catch unanswered questions and review decisions, owners, and source quotes. You start it on your Mac.

Free plan — no credit card required

No bot joins your call

Turn the conversation into a decision you can return to.

A decision you can trace

Decision → owner → source quote

Illustrative walkthrough
Cruxwing product demo

A copilot for the whole conversation.

Bring your context, stay caught up while you talk, and turn the next steps into action.

  1. Private capture Keep transcription on your Mac
    Capture optionsCapture Zoom, Meet or Teams without a bot. Connect LM Studio for assistant answers from a local model.
  2. Connected context Bring the context that matters
    Choose your contextSet your goal and role. Add past calls and choose context from connected apps: Notion, Linear, Fireflies and more.
  3. Live summary Catch up without asking
    How it updatesA summary refreshes during the call, highlighting what changed, why it matters to your goal and what to do next.
  4. Blind spots + assistant Catch the gap. Talk it through.
    From finding to follow-upBlind-spot scans surface missed risks with transcript evidence. Cruxwing automatically selects relevant skills from the library for your goal and conversation. Explore findings with the assistant.
  5. Exports Put the next steps to work
    Review before sendingReview and file tasks to a connected tracker, or export answers to Google Docs and Notion. You choose what gets sent.

Compare the help you get on the call.

Choose your current tool and the work you do. Compare capture, live help, and the evidence you can return to.

Full comparison
Structural comparison of Cruxwing, Granola, Krisp, Otter, Fireflies, Read AI and Grain
  Cruxwing Granola Krisp Otter Fireflies Read AI Grain
No bot joins YesCapture starts from your Mac YesCaptures computer and microphone audio YesBot-free capture Desktop optionRecord without a meeting bot Desktop optionTake Notes uses system audio Desktop optionRecord audio directly from your computer Desktop optionCapture without a meeting bot
On-device transcription Available on MacChoose the on-device engine in Settings CloudUses transcription providers Enterprise optionPrivate on-device transcription CloudInternet needed for live transcription CloudStored and processed in cloud infrastructure Not documentedDesktop capture is not a local-processing guarantee Not documentedLocal processing not specified in live guide
Decision tracing Decision + owner + quoteReview the evidence, then confirm the outcome Search decisionsAsk Chat for specific decisions or details Action itemsMeeting notes highlight next steps Source-linked answersAnswers link back to source conversations Live action itemsRequest tasks, next steps and decisions Meeting insightsAsk Read about meeting content Linked timestampsNotes and action items link to the recording moment
Blind-spot scan Transcript-cited gapsUnprompted; every card must quote the transcript line that triggered it On-demand chatAsk about gaps and decisions AI ChatAsk follow-up questions On-demand chatAsk questions during the call Sales AssistLive sales suggestions require its meeting bot Ask ReadMeeting insights on request Not documentedLive guide covers transcript and notes
Live help mid-call AI advisorAsk a question using the call and your chosen context During-call chatRequest recaps and explanations Call Center AI optionLive Knowledge Chat; transcript access requires enablement During-call chatAsk questions in a live meeting Live AssistAskFred answers while you talk During-call Ask ReadCatch up and pull context from past calls Not documentedLive guide covers transcript and notes
Connect a local model for AI answers LM StudioAssistant answers on your Mac Not supportedModels provided by Granola Not documentedLocal audio processing is separate Not documentedHosted AI Chat models Not documentedCloud processing documented Not documentedRead models and hosted LLMs Not documentedMeeting data via hosted AI clients
Live summary — while the call is still going AutomaticRefreshes during the call, focused on your goal On demandOpen Chat and request a recap After callLive transcript; summary after AutomaticLive summary in Outline On-demand recapClick Summarize So Far; separate AI Notes update automatically In live dashboardOpen the dashboard for automatic summary updates After callLive transcript and notes; summary on meeting page

Checked against linked vendor documentation on 8 September 2026. “Not documented” means the reviewed guide does not establish that feature; it does not mean the feature is unavailable. Bot-free capture does not by itself mean on-device processing. Plan and platform limits may apply.

What do unclear decisions cost each week?

Estimate the time spent reopening decisions and confirming who owns the next step.

1–1,000 people revisiting each decision.
$0–1,000 per person, in USD.
0–100 calls for this team.

Share of revisiting time you think Cruxwing could help you avoid (10–90%). A modeled assumption, not an industry benchmark or measured product result.

Adjust the assumptions

This scenario assumes everyone entered revisits one decision after each call.

0–60 minutes per person, per call.

Weekly team hours = people × calls × revisit minutes ÷ 60. Cost = hours × hourly rate. Potential hours recovered = hours × your recovery assumption.

modeled team hours revisiting / week
estimated cost / week in decision drift
potential hours recovered / week with Cruxwing

Illustrative scenario, not measured product savings. Results depend on your assumptions and how your team uses Cruxwing. Your inputs stay in this calculator.

Download free for Mac

Free plan. No credit card.

Unlimited private meeting memory. Pay for live judgment and managed AI.

Capture stays unlimited on-device. Paid plans add live help, and more blind-spot scans and assistant prompts for follow-ups and briefs, with hard monthly limits.

Blind-spot scan

One pass the assistant makes over the conversation on its own, looking for what was not said — a missed objection, a claim nobody checked. It quotes the transcript line that proves it.

Assistant prompt

One thing you ask it, typed or by button — mid-call or afterwards. Counted separately from scans, so a long meeting watching itself cannot eat the answers you saved for later.

Live help

Scans and answers while the call is still running, not in tomorrow’s recap. Recording and on-device transcription are unmetered on every plan and draw on neither.

Free

Private capture and a real product, not a trial.

$0 / mo
  • Unlimited on-device transcription, kept private — no meeting cap, no trial clockUnlimited on-device transcription
  • Search every past call by what was said, and get the quote back, not a summarySearch past calls with source quotes
  • 10 live blind-spot scans and 10 assistant answers a month10 blind-spot scans + 10 prompt units / month
What do I get?

Private capture and a real product, not a trial.

Unlimited on-device transcription, kept private — no meeting cap, no trial clock

Search every past call by what was said, and get the quote back, not a summary

10 live blind-spot scans and 10 assistant answers a month

Blind-spot scans per month
10
Assistant prompt units per month
10

Scans flag what may be missing. Use assistant prompts for live help and follow-ups. Longer-context requests use two prompt units.

Unlimited recording and on-device transcription.

Monthly allowances apply when signed in. Guests start with 10 scans and 10 prompt units.

Start free

Recommended

Live Copilot

For people who want the sharp question while it still matters.

$19 / user / mo

3-day free trial

  • Transcript-cited blind spots while you can still act, not in tomorrow’s recapTranscript-cited blind spots
  • Follow-up briefs from human-confirmed outcomes, each carrying the line that decided itBriefs from confirmed decisions
  • Your connected tools searched mid-call, so «did we agree this already?» has an answerSearch connected tools during calls

350 blind-spot scans + 50 prompt units / month

What do I get?

For people who want the sharp question while it still matters.

Transcript-cited blind spots while you can still act, not in tomorrow’s recap

Follow-up briefs from human-confirmed outcomes, each carrying the line that decided it

Your connected tools searched mid-call, so «did we agree this already?» has an answer

Blind-spot scans per month
350
Assistant prompt units per month
50

Scans flag what may be missing. Use assistant prompts for live help and follow-ups. Longer-context requests use two prompt units.

Unlimited recording and on-device transcription.

The same monthly allowances apply with Annual billing.

Pay

Power

For high-stakes calls and heavier reasoning.

$49 / user / mo

3-day free trial

  • Deeper reasoning and multi-model review on demand, for the calls that decide budgetsDeeper reasoning on demand
  • Room for a full week of briefs and follow-ups before the paid AI pausesMore room for briefs and follow-ups
  • The hard cap stays on: when it pauses, recording and on-device transcription do notAI pauses at your cap; recording continues

800 blind-spot scans + 100 prompt units / month

What do I get?

For high-stakes calls and heavier reasoning.

Deeper reasoning and multi-model review on demand, for the calls that decide budgets

Room for a full week of briefs and follow-ups before the paid AI pauses

The hard cap stays on: when it pauses, recording and on-device transcription do not

Blind-spot scans per month
800
Assistant prompt units per month
100

Scans flag what may be missing. Use assistant prompts for live help and follow-ups. Longer-context requests use two prompt units.

Unlimited recording and on-device transcription.

The same monthly allowances apply with Annual billing.

Pay

High-volume

In calls all day? For back-to-back schedules.

$99 / user / mo

3-day free trial

  • 1,400 blind-spot scans and 200 assistant prompts a month — the most of either we sell1,400 blind-spot scans + 200 prompt units / month
  • Room for briefs and follow-ups across a full week of back-to-back meetingsEverything in Power
  • Everything in Power, with the same hard cap: the paid AI pauses, recording never doesHigher capacity, same models as Power

Same models and same answers as Power — you are buying capacity, not better judgment.

What do I get?

In calls all day? For back-to-back schedules.

1,400 blind-spot scans and 200 assistant prompts a month — the most of either we sell

Room for briefs and follow-ups across a full week of back-to-back meetings

Everything in Power, with the same hard cap: the paid AI pauses, recording never does

Same models and same answers as Power — you are buying capacity, not better judgment.

Blind-spot scans per month
1,400
Assistant prompt units per month
200

Scans flag what may be missing. Use assistant prompts for live help and follow-ups. Longer-context requests use two prompt units.

Unlimited recording and on-device transcription.

The same monthly allowances apply with Annual billing.

Pay

Same call. Different Friday.

An illustrative example. You review and confirm what goes into the decision log.

Without Cruxwing

By Friday: what did we agree, and who owns it?

See the example
  1. Monday · on the callDecision made: send the revised scope.
  2. Friday · in the team chat“What did we agree — and who was sending it?”
  3. The costReopen the discussion. Find the context again.

With Cruxwing

Decision logged. Owner named. Quote captured.

See the example
  1. Decision logged · after your reviewSend the revised scope by Friday.
  2. Owner namedMaya — confirmed from the conversation.
  3. Quote capturedReturn to the exact line behind the decision.
“I’ll send the revised scope by Friday.” — Maya

More blind-spot scans

  • 100 scansfor $7
  • 300 scansfor $18
  • 1,000 scansfor $59

Same background watch, when a heavy month outruns the plan.

More assistant prompts

  • 50 promptsfor $3
  • 150 promptsfor $6
  • 500 promptsfor $18

Any model you can reach, one prompt each – a very large prompt counts as two.

No surprise overages. Hard monthly caps are on by default. When a balance runs out, the paid AI pauses; recording and on-device transcription continue. There are no automatic charges or usage-based overages.

Questions

The ten things people ask before they buy.

Blind-spot scans and assistant prompts — how do they count?

Two things, and you can count both. A blind-spot scan is one pass the assistant makes over the conversation on its own, looking for what was not said — a missed objection, a claim nobody checked. An assistant prompt is one thing you ask it, typed or by button.

They are separate so a long meeting watching itself cannot eat the answers you saved for afterwards. Recording and on-device transcription are unmetered and draw on neither.

Does it work offline?

Capture and transcription do. Cruxwing records system audio and your mic natively and transcribes on-device with Whisper sized to your chip, so a flight or a locked-down network changes nothing about getting the transcript.

Live blind spots and managed AI actions need a connection. They resume when you are back online, against the transcript captured without them.

How is this different from Granola, Fireflies or Otter?

Not by capture any more — Granola, Otter and Fireflies all record bot-free now, and all three answer questions mid-call if you ask. Their bet is the archive.

Cruxwing bets on the room, while the decision is still reversible. It speaks first — a blind spot that has to quote your transcript — and writes what the team confirmed into an append-only ledger. It is also the only one here that transcribes fully on-device. Row by row, including the rows Cruxwing loses: the comparison table.

How does this compare to Krisp?

Krisp wins on price and on habit: $16 a month, or $8 billed annually, against our $19, plus noise cancellation — a reason to open an app daily that we have no equivalent for. Both of us run on Mac and Windows.

Two rows go the other way. Krisp puts on-device transcription behind Enterprise; ours is free, unlimited and offline — on a Mac, which is the honest limit. And Krisp writes notes, not an append-only ledger of what a human confirmed. Decide on that; on notes the two are close.

Does Cruxwing translate meetings into another language?

No. It transcribes in the language being spoken — you pick that language — but it does not translate speech, produce translated audio, or convert accents.

Does my meeting data go to the cloud?

That is your choice, per meeting. Choose on-device transcription and the audio never leaves your Mac. Choose a managed engine and audio is streamed to that provider for processing; the privacy policy names every sub-processor, what it receives, and where it is.

AI answers go to the model provider you select; the privacy policy names those providers and what they receive. Sign-in credentials are sealed in the macOS Keychain.

What happens when they run out?

Nothing stops recording. Recording and on-device transcription carry on, because on-device capture is unmetered on every plan. When your blind-spot scans are used up the assistant stops watching on its own; when your assistant prompts are used up it stops answering. The two run out independently.

The app is told once and stops asking, so background watching goes quiet instead of retrying against an empty allowance. You get one plain message naming which of the two ran out.

How do you stop it inventing things?

Every blind spot has to quote your transcript, and the quote is checked against what was actually recognised — if it does not appear, the card is discarded rather than shown to you unsourced. A model that paraphrases instead of quoting produces nothing.

Structured output gets a second gate: owners and dates are checked back against the transcript, and an unverifiable owner becomes an open question rather than a name. That is also why a quiet meeting produces a short answer — silence is a real result.

What if I don't use my whole monthly limit?

Each billing period starts with your plan's full allowance, and the floor never drops to nothing: on-device transcription is unlimited on every plan, including Free.

Annual plans save two months: Live Copilot is $190, Power is $490, and High-volume is $990, each equal to ten monthly payments.

Does it only work for meetings?

No. Pick the recording type — meeting, lecture, podcast, tutorial, interview, presentation, or your own label — and the output contract changes with it. A lecture comes back as concepts, evidence and open questions rather than a decision log.

On automatic it reads the visible window titles and proposes a type; your manual choice overrides it. The ledger never changes — a lecture takeaway is never filed as a team decision. the recording type sets what changes.

See it on your own meeting.

60 minutes · Your call · Your stack

The calendar loads from Calendly as you reach this section, sharing your IP address and browser details. Booking sends your name and email to Calendly, our scheduling processor, to create the invite. Privacy policy.

Available times load here as you reach this section.

Open calendar in a new tab