Best AI Meeting Assistants in 2026, Ranked Honestly

TL;DR: The best AI meeting assistants in 2026 split along two axes: whether a bot joins your call, and whether audio leaves your machine. Everpaper is our pick for managers on Apple Silicon who want on-device transcription, hidden-speaker options, and no bot in the room. Granola is the strongest bot-free cloud option. Otter.ai and Fireflies.ai are the bot-based incumbents with deep integrations. ScreenPipe is the developer's local-first recorder.

Bias declared, once: Everpaper is our product. Its on-device architecture is why we rank it first for privacy-first Apple Silicon Mac users. Read that judgment with our interest in mind; the drawbacks below use the same standard as every competitor.

How we evaluated

We scored each assistant on five criteria, weighted for managers and consultants rather than enterprise IT:

• Capture model. Does a visible bot join the call, or does the app record locally? Bots announce themselves in the participant list, get kicked from calls, and two tools on this list face litigation. Local capture works on any platform and never asks permission from the other side of the table, which is also its ethical burden; we'll come back to that.
• Data path. Where does the audio go, who transcribes it, and can any of it train someone else's model?
• Transcript quality and speaker attribution. A summary you can't trace back to who said what is a liability in a client engagement.
• What you keep. Recording, playback, export formats, integrations.
• Cost of ownership. Not just the sticker price: minute caps, seat counts, and whether you're renting the tool forever.

Published competitor prices below were checked on 2026-08-04 against vendor pricing pages.

1. Everpaper: best on-device meeting assistant for Mac

Overview

Everpaper is a private, on-device daybook for macOS, and its meeting pillar, called The Minutes, records and transcribes meetings entirely on your machine. The other two pillars, the Day Journal and the Second Brain, sit beside it, but this article is about meetings. No bot joins the call. No audio leaves the Mac for transcription. It recognizes Zoom, Google Meet, Microsoft Teams, Webex, Around, and Whereby links, and you start a recording with ⇧⌘M, the Record tile, the menu-bar item, or Index (⌘K) → Begin minutes.

What we like

The transcription runs on the Neural Engine using Parakeet (parakeet-tdt-0.6b-v3, the default engine), with Whisper and MOSS (MOSS-Transcribe-Diarize, transcription and diarization in one model, Apache 2.0, ~2 GB download) as alternatives. All of it local. While the meeting runs, a floating panel shows level meters and a live speaker-labeled transcript, with tabs for Notes (⌘1–⌘3), slash snippets like /decision and /action, and an Ask tab for AI over the live transcript with quick prompts like Catch Up and Challenge.

Two features matter specifically for managers. First, mic and system audio are recorded as separate time-aligned tracks; your voice and the room's voice never get flattened into one stream. Second, Hide my voice / Hide any speaker (⌘4, persists across meetings) removes a speaker from the record entirely, which is the cleanest answer we know of to "my counsel shouldn't be on tape here."

Calendar Auto-Start is opt-in with read-only local calendar access; events are never uploaded. Set it to Start recording automatically and you get a 5-second cancellable countdown toast showing the meeting title, service, and attendee count; Return starts, Esc cancels. All-day and declined events are always skipped, and it never auto-records a meeting you didn't answer.

After the meeting, Extract todos pulls action items into Today's Focus automatically, and prompt templates (Summary, Action Items & Decisions) generate follow-ups. Everything exports to txt, markdown, srt, vtt, or json via the bundled everpaper-cli (see the CLI docs), which can also transcribe any audio or video file on-device. Export details live in the export docs.

Drawbacks

Honest list, because a privacy product caught overstating loses its one asset: Everpaper is macOS 14.2+ only, Apple Silicon only, so Intel Macs are unsupported and there's no iOS, Windows, or web version. Access is limited during the private beta. The AI features (meeting Ask, summaries) wait until you configure a provider, though local options exist (Ollama, LM Studio, oMLX, Apple Foundation Models) so nothing has to leave the Mac even then. And MOSS is happiest on 16 GB machines.

Pricing

Everpaper is in private beta with limited spots, with a founding-license option for early supporters, not a monthly subscription. No public price as of 2026-08-04.

Best for

Managers and consultants on Apple Silicon Macs who bill by the hour and can't afford a data incident (client calls, board prep, HR conversations) and who'd rather own a tool than rent one.

2. Granola: best bot-free cloud option

Overview

Granola is the bot-free notepad that proved you don't need a participant-slot bot to get good meeting notes. The desktop app (macOS and Windows, plus iOS and Android apps for in-person meetings) captures microphone and system audio locally, streams it to transcription providers, and merges your typed notes with the transcript into polished AI notes. There is no meeting bot; Granola runs only on your computer and never auto-transcribes without you clicking into a note.

What we like

The note quality is widely praised; Nat Friedman called it "the best meeting notes of anything I've tried," and a long-term user reported that it replaced their note-taking workflow. The human-in-the-loop design is useful: you jot rough notes during the call, and the AI fleshes them out, which gives you more control than a fully auto-generated summary. It is also platform-agnostic on desktop because anything playing through system audio works. Business includes MCP and API access, while mobile apps cover in-person meetings. Granola is SOC 2 Type 2 certified and publishes public security post-mortems.

Drawbacks

Audio goes to the cloud for transcription (Deepgram and AssemblyAI) and summarization (OpenAI and Anthropic), with notes stored in US-hosted AWS. Audio itself isn't stored, only the transcript, but that also means no playback, ever: you cannot verify a quote against the recording, which stings in a dispute. Speaker attribution is weak: desktop transcripts default to "Me" (green, your mic) and "Them" (grey, system audio). Named speaker tags are limited to Google Meet through a browser extension and Zoom on macOS, not Teams. Granola trains its own models on anonymized user data by default on consumer tiers; you can opt out in Settings, and it's off by default on Enterprise.

Pricing

Checked 2026-08-04 (per user per month, USD): Basic free; the pricing page says "limited meeting history" but does not publish the exact cap. Business $14 (unlimited history, integrations with Notion, Slack, HubSpot, Attio, Affinity, Zapier, MCP, API); Enterprise $35 (SSO, admin controls, org-wide auto-deletion, model training off for the team).

Best for

Consultants and managers who want the best-written notes in the category, accept the cloud data path, and mostly live on Zoom and Google Meet.

3. Otter.ai: best for CRM-integrated teams

Overview

Otter.ai is the category's incumbent: a cloud meeting bot (OtterPilot) that joins Zoom, Google Meet, and Microsoft Teams as a visible participant, records and transcribes in real time, captures slides, and emails a summary afterward. It also offers desktop and mobile apps, a Chrome extension, and file import, but the bot is the product's center of gravity, auto-joining via calendar sync.

What we like

The live transcription with real-time editing, highlighting, and commenting is mature, and the cross-meeting AI Chat, which asks questions across your entire meeting archive, is something bot-free tools can't easily match. The integration depth is the real story: Salesforce, HubSpot, Zapier from Pro up, and an Enterprise tier with a Sales Notetaker, SSO/SCIM, API and webhooks, and a HIPAA add-on. If your org measures meetings in pipeline terms, Otter is built for that.

Drawbacks

The bot is also the liability. Otter faces a consolidated federal class action, In re Otter.AI Privacy Litigation (N.D. Cal.), alleging recording without all-party consent and using recordings to train AI; a motion-to-dismiss hearing was held May 20, 2026. The privacy policy (effective June 16, 2026) explicitly reserves the right to train "proprietary AI technology on de-identified audio recordings and on transcriptions." Users report the bot joining confidential meetings uninvited, spam behavior that emails whole companies, hit-or-miss accuracy with "Speaker 1/2" misattribution needing manual fixes, and minute-cap stress: the free tier is 300 minutes a month with a 30-minute per-conversation cap, and deleted conversations still count.

Pricing

Checked 2026-08-04 (per user per month, USD): Basic free (300 monthly transcription minutes, 30 min per conversation, 3 lifetime file imports); Pro $16.99 monthly or $8.49 billed annual (1,200 minutes/mo, 90 min per meeting); Business $30 monthly or $24 billed annual (4 hours per meeting, admin features, 3 concurrent meetings); Enterprise custom.

Best for

Sales and success teams already living in Salesforce or HubSpot who want meetings piped into the CRM and accept the bot's visibility and the legal cloud around consent.

4. Fireflies.ai: best for multilingual conversation analytics

Overview

Fireflies.ai sends its bot "Fred" into your Zoom, Meet, Teams, or Webex calls, transcribes in 100+ languages, and layers conversation intelligence and team analytics on top. It's the most team-oriented tool in this list, a workflow hub with an API on every tier and unlimited integrations from Pro up.

What we like

The free tier is one of the more usable ones around: unlimited transcription (rate-limited) and AI summaries, 400 minutes of storage per team, Zoom/Meet/Teams support, the AskFred assistant, file upload, and API access. The language coverage, 100+ languages with a multi-language mode on Business, beats every other tool here. And the enterprise posture is stronger than most consumer notetakers: SOC 2 Type II, GDPR, HIPAA with a BAA on Enterprise, SSO/SCIM, private storage, and a stated policy of not training on your data by default.

Drawbacks

The bot's auto-join behavior is on by default via calendar sync, and it's the source of the most complaints: Fred shows up uninvited in sensitive calls and users report it's hard to fully disable. Fireflies faces a BIPA class action, Cruz v. Fireflies.AI Corp., filed December 2025 in the Central District of Illinois, alleging it collected and stored voiceprints of meeting participants without the notice and written consent Illinois law requires. Reviews flag accuracy drops with heavy accents and crosstalk, and Trustpilot users complain about AI-credit billing confusion and higher-than-expected bills.

Pricing

Checked 2026-08-04 (per seat per month, USD): Free $0 (unlimited transcription, 400 mins storage per team, 20 AI credits); Pro $10 billed annual or $18 monthly (8,000 mins storage per seat, video recording, downloads); Business $19 annual or $29 monthly (unlimited storage, conversation intelligence, team analytics); Enterprise $39, annual only (SSO/SCIM, HIPAA BAA, private storage, rules engine).

Best for

Multinational teams who need 100+ language coverage and conversation analytics, and whose compliance teams are satisfied with a cloud vendor holding recordings, BIPA suit notwithstanding.

5. ScreenPipe: best local-first recorder for developers

Overview

ScreenPipe isn't really a meeting assistant. It's a 24/7 local recorder for your screen and microphone (macOS, Windows, Linux, written in Rust), with a local search API and scheduled AI agents called "pipes" on top. It's in this list because it offers no-bot, local-first capture: it records whatever your machine displays and hears, so it can capture meetings on any platform without adding a participant to the call.

What we like

The local-first claim is real: data lives in ~/.screenpipe/db.sqlite, capture and OCR and local Whisper transcription all work offline, and there's a default-on PII redaction flag (--use-pii-removal) that scrubs emails, phone numbers, and card numbers before storage. The developer surface is genuinely good: a local HTTP API at localhost:3030, an MCP server for Claude Desktop, Cursor, and VS Code, and pipes defined as a pipe.md prompt plus a schedule (meeting-ended trigger → summary → Obsidian note), runnable on local models via Ollama. Y Combinator backed (S26).

Drawbacks

It's infrastructure, not a product a manager can hand to their EA. ScreenPipe's official FAQ does not document speaker identification or diarization. The official troubleshooting docs tell users to lower capture FPS and use whisper-tiny on older machines, an acknowledgment that 24/7 capture costs CPU and battery. The business model has confused its own community since 2024, and the June 2026 license move from MIT to a source-available commercial license for business use stung. The official platform list names macOS, Windows, and Linux; it does not list mobile apps.

Pricing

Checked 2026-08-04 against the supplied ScreenPipe pricing page: Free $0 (7-day Business trial, local capture, AI search, meeting notes, 10 AI credits/month); Basic $25/month or $250/year (150 credits, efficient models, unlimited user-installed scheduled tasks and history); Business $50/seat/month or $500/seat/year (400 credits, frontier Claude and GPT models, cloud sync, automations, and team workspace).

Best for

Developers and technical founders who want to build their own meeting-memory pipeline on local data, and who'll happily trade polish for an API.

Honorable mentions

Fathom and Read.ai.

Limitless is not a current new-customer recommendation. Its official site says the Pendant is no longer sold and desktop and web recording are disabled. Existing Pendant customers receive Unlimited free with support promised throughout 2026; the site does not publish a purchasable new-customer plan.

How to choose

Start with the two axes from the TL;DR table, because they eliminate most of the field:

Bot or no bot? If your meetings include clients, counsel, HR, or anyone who'd flinch at "OtterPilot has joined," a bot-based tool costs you social capital every call, and both bot vendors here face active litigation over consent and voiceprints. No-bot tools (Everpaper, Granola, ScreenPipe) record from your own machine. That power cuts both ways: recording a call without telling anyone may be legal where you sit and illegal where your counterparty sits. Know your jurisdiction's consent rules, and when in doubt, say you're recording.

Cloud or on-device? If the audio can never leave the machine (client confidentiality, regulated work, or just principle), the field narrows to Everpaper and ScreenPipe, and ScreenPipe is a developer toolkit. Granola and the bot tools all send audio or transcripts to US clouds; read their training-data policies (Granola trains on anonymized data by default on consumer tiers; Otter reserves the right to train on de-identified recordings; Fireflies says it doesn't train by default).

Then the practical filters:

• On Windows, or need mobile? Everpaper is out (macOS 14.2+, Apple Silicon only). Granola or Fireflies.
• Need CRM piping? Otter (Salesforce/HubSpot) or Fireflies (unlimited integrations from Pro).
• Need 100+ languages? Fireflies, clearly.
• Want a $0 local path? ScreenPipe's Free plan. Want a non-subscription license? Everpaper's founding license.
• Want searchable recall across everything, not just meetings? Everpaper's Ask the Book answers from your captured days with per-frame sources.
• Want the best-written notes and accept the cloud? Granola.

For a deeper look at Everpaper's meeting capture, see the Minutes documentation, and for the data-path details behind every claim above, the privacy model docs. If you're weighing a specific rival, our head-to-head comparison pages (linked from the blog index) go deeper per tool.

FAQ

What is the best AI meeting assistant in 2026?

For privacy-first managers on Apple Silicon Macs, Everpaper: on-device transcription, no bot, separate mic/system tracks, speaker hiding. For the best-written notes in the cloud, Granola. For CRM-integrated teams, Otter.ai. For multilingual analytics, Fireflies.ai. For developers, ScreenPipe.

Do AI meeting bots record without consent?

Bot-based tools like Otter.ai's OtterPilot and Fireflies' Fred join calls as visible participants and record when the meeting host or an auto-join setting allows it. Legal consent depends on jurisdiction and the facts of the recording. As of 2026-08-04, the pending Otter suit alleges recording without all-party consent, while the Fireflies suit alleges voiceprint collection without BIPA notice and written consent; neither allegation is a finding of liability.

Which AI meeting assistants don't use a bot?

Everpaper, Granola, and ScreenPipe capture microphone or system audio from the user's machine, so no participant bot joins the call. Everpaper transcribes locally. ScreenPipe can transcribe with local Whisper. Granola streams audio to cloud transcription providers. Otter.ai and Fireflies.ai are bot-based, though both also offer desktop or mobile recording modes.

Is there an AI meeting assistant that works offline?

Everpaper transcribes on-device via Parakeet, Whisper, or MOSS on the Neural Engine; capture, transcription, storage, and search never need a network. ScreenPipe also records, transcribes (local Whisper), and searches fully offline. Granola, Otter, and Fireflies all require connectivity for transcription and AI features.

What's the cheapest AI meeting assistant?

Fireflies' free tier (unlimited transcription, 400 mins storage) and Otter's free tier (300 mins/mo) are the most usable $0 options among the cloud tools. Granola's pricing page labels Basic meeting history "limited" without publishing the cap. ScreenPipe's CLI is free. Everpaper is in private beta with a founding-license model, not a subscription, with no public price as of 2026-08-04.

Can AI meeting assistants identify who said what?

Everpaper supports live speaker labels, and its opt-in MOSS model handles transcription and diarization together on-device. Otter and Fireflies document cloud speaker identification or diarization, though reviews report misattribution. Granola defaults to "Me"/"Them," with named speaker tags limited to Google Meet through an extension and Zoom on macOS. ScreenPipe's official FAQ does not document speaker identification or diarization.

Methodology

We evaluated the five ranked tools against their official documentation, pricing pages, security pages, and privacy policies on 2026-08-04, plus third-party reviews (G2, Capterra, Trustpilot, tl;dv, Zapier) and court reporting where litigation is on file. Published price amounts were checked against vendor pages or supplied vendor pricing. Granola's page does not publish the Basic history cap. Everpaper is our product and is ranked first on the strength of its on-device architecture; its limitations (macOS-only, Apple Silicon only, private beta, no mobile or web) are stated in full. We did not test with live client calls, and we don't accept placement fees; nobody here paid to be ranked.

---

Everpaper is in private beta with limited spots: on-device meeting transcription, a day journal, and a second brain, all kept on your Mac. Join the waitlist at everpaper.app, or read the docs on meetings and privacy first.