Back to Blog
16 min read

AI Conversation Memory Tools in 2026: A Complete Comparison

Short answer for 2026: there's no single best — there are best-for-X answers. For ambient capture, Omi ($89, open-source). For meetings, Granola or Plaud (software-only beats a wearable for this use case). For existing-pendant intelligence on Limitless or Omi, Fluent. For a companion AI (not memory), Friend or Pi. Skip the rest until they prove themselves.

The category that didn't exist in 2022 has become genuinely confusing in 2026.

There are pendants you wear on your collar. Software that listens through your laptop. Hybrid systems that capture phone calls. "Memory" apps that mean something different in every product description. Three of the loudest brands have been acquired in the past twelve months. New entrants ship monthly. The space is real, the products are mostly good, and choosing between them has gotten harder, not easier.

This guide is the comparison page we wished existed when we were trying to understand the market ourselves. Every tool worth considering, the dimensions that matter, and the questions you should ask before committing — including the ones nobody else is asking.

A note on bias: Fluent is one of the tools listed below. We've tried to be ruthlessly fair anyway. When something else is the right answer, we say so.

The headline comparison

ToolTypePricePlatformPrivacyBest forStatus
LimitlessHardware pendantSales haltediOS, Mac, WebCloudAmbient captureAcquired by Meta (Dec 2025)
OmiHardware pendant~$89iOS, AndroidOpen-source, self-hostablePrivacy-first captureActive, independent
Plaud (Note / NotePin)Hardware pendantMid-rangeiOS, Android, WebCloudMeeting captureActive, independent
BeeHardware pendant~$49iOS, AndroidCloudCasual consumer useAcquired by Amazon
FriendHardware pendant~$129iOSCloudCompanion (not memory)Active
RewindSoftware (Mac)SunsetMacLocal-first (legacy)Screen memoryFolded into Limitless
GranolaSoftware (Mac)~$18/moMacCloudWork meetingsActive
OtterSoftware$17–30/moWeb, iOS, AndroidCloudB2B meetingsActive, public
Personal.aiSoftware$40/mo+Web, iOSCloudPersonal AI memory stackActive
ScreenpipeSoftware (open-source)Free / self-hostMac, Windows, LinuxLocal-onlyPower-user screen memoryActive, independent
FluentSoftware (iOS)$9.99–19.99/moiOS; integrates with Limitless + OmiEncrypted at rest (per-user keys on roadmap)Transcripts + summaries + person profiles + agentic calendar + MCP integrationActive, independent

Notes on the table. Sales halted means the company stopped selling new hardware; units already in the wild still work. Integrates with Limitless + Omi means Fluent ingests audio directly from those pendants today; more pendant integrations are on the roadmap. Fluent does not capture audio without a pendant. Encrypted at rest means data is protected on disk via service-managed keys. Per-user encryption — where the company literally cannot decrypt your data even if subpoenaed — is on the roadmap, not shipped.

Skip ahead by use case

Hardware pendants

Five products in this category right now. Three of the five have been acquired in the past twelve months. We'll come back to that.

Limitless — orphaned

Acquired by Meta in December 2025. Pendant sales halted. Existing units still pair and capture, but the long-term path is unclear. The Mac and web products are sunsetting in stages.

If you own a Limitless pendant, the device hasn't stopped working. But the company that supports it is now a feature team inside the world's largest social network, and the product roadmap is no longer transparent. Most active Limitless users are quietly migrating their workflow to other tools while keeping the pendant as a capture device.

What to do if you have one: export your data while the export tools still work, pair the pendant with software that isn't going anywhere (more on this below), and don't rush a hardware replacement. (Full deep-dive on the orphaned-Limitless decision: Limitless Pendant Alternatives in 2026. For getting actual use out of the pendant you already own: What to Do With the AI Pendant You Bought.)

Omi

About $89. Open-source firmware, self-hostable backend. iOS and Android.

Omi (from Based Hardware) is the closest hardware replacement for Limitless if your reason for wearing one was "I want a pendant." Smaller, cheaper, firmware on GitHub. A meaningful number of users self-host the entire stack on their own infrastructure.

The trade is rough edges. Battery sometimes dies, Bluetooth sometimes doesn't pair, the app feels hobbyist-tier compared to what Plaud or Limitless ship. Setup expects you to be comfortable with the kind of fiddly things you'd accept from a project on GitHub. If you found Limitless's UX rough, Omi will not feel better.

Right answer for engineers, privacy researchers, self-hosters, and anyone who would rather debug a pairing issue than accept a vendor lock-in.

Plaud (Note and NotePin)

Mid-range hardware. Business-leaning. iOS, Android, web.

Plaud is the most finished pendant on the market and the SEO juggernaut of the category. They've thrown ten times more content marketing at this space than anyone else, which is partly why you've heard of them. Long battery, magnetic mount, native phone-call recording, summaries that are noticeably better than Limitless's were at sunset.

Cloud-default. Your conversations live on Plaud's servers, in some configuration of access controls you'll never read. The privacy posture isn't adversarial. It isn't Omi.

The product is shaped for the meeting. Not the dinner conversation, the walk, the call from your kid's school. Right answer for professionals who mostly capture meetings, don't mind the cloud trade, and remember to press the button.

Bee

About $49. Cloud-default. iOS and Android. Now Amazon-owned.

Bee is the entry point. Forty-nine dollars for a pendant whose entire job is to listen to you and report back. Capture quality is fine. The app is consumer-shaped. The integration story is whatever Amazon decides on the morning of the all-hands.

If "what happens to my data" is a question you'd like to be able to answer, you can't. But you only paid $49, so the bar was low to begin with. Same acquisition pattern as Limitless, just a different acquirer.

Right answer for testing the category without thinking too hard about it.

Friend

About $129. Companion-shaped. iOS only.

Friend is on this list to be excluded. The marketing positions it as your friend who's always with you, which is either soothing or alarming depending on the rest of your life. If a friend was what you secretly wanted from Limitless, Friend is honest about that. If you wanted memory, pattern recognition, or anything that resembles work output, you're shopping in the wrong aisle.

Software for meetings

If you don't actually need ambient daily-life capture and "meetings" is your real use case, the answer is probably software, not hardware.

Granola — the Mac meeting incumbent

About $18/mo. Mac-only. Cloud-stored, no-bot-in-meeting model.

Granola listens through your laptop's mic during meetings. No bot joins the call. No Zoom calendar integration required. Meeting summaries are sharp, the UI is fast, and the brand is built around "we don't crash your meeting."

Mac only. Built for the meeting, not the conversation life around it. Anything that happens away from your laptop is invisible to it. Cloud storage of summaries.

Fluent and Granola serve different needs and don't really overlap. Granola needs a Mac with a meeting. Fluent needs a Limitless or Omi pendant for ambient capture. If your conversation life is mostly Zoom calls, Granola. If it's mostly walks, dinners, and wearing a pendant, Fluent. If it's both, you might want both.

Otter — the cloud incumbent

$17–$30/mo for consumer plans, more for teams. Web, iOS, Android.

Otter is the oldest brand in this category, with the largest install base and the most institutional muscle. It's built for teams: bot joins your meeting, transcript is shared, action items are extracted, admin tools control retention. Mature product. Strong B2B integrations (Zoom, Teams, Slack). Generous free tier.

Designed for teams, not individuals. The bot-joins-the-meeting model is exactly what Granola was built to avoid. The privacy posture assumes you're an org with admin policies.

Right answer for B2B teams. Less right for personal use.

Privacy-first

If your top priority is nobody but me can read this, the field narrows fast.

OptionWhat you getTrade-off
Omi (self-hosted)Open-source hardware + your own infrastructureYou have to run servers
ScreenpipeOpen-source local screen capture, runs entirely on your machineMac/Linux only, screen-only (no audio capture from in-person conversations)
FluentEncrypted at rest; per-user encryption on roadmapToday the service can decrypt; full user-key encryption hasn't shipped yet

True local-only with audio capture is genuinely hard right now. Omi self-hosted is the closest thing to "your conversations never touch a cloud you don't own." Everything else, Fluent included, is some flavor of trust-the-vendor or trust-the-encryption-at-rest.

Budget pick

If you just want to try the category for under $100 total: Bee ($49) is the cheapest hardware on-ramp, and the Fluent free trial is the cheapest software on-ramp. There's no need to combine them. Pick one, see if you actually use it for two weeks, then make a real decision.

Prosumer

If you're a solo founder, executive coach, consultant, sales rep, or manager running recurring 1:1s — your use case is different from both consumer and B2B-team.

You need conversation memory across contexts, not just inside scheduled meetings. Clients call you on the phone. You have lunch meetings. You debrief in cars. You need patterns over time, not just per-meeting summaries — what did this client mention three meetings ago that you forgot to follow up on. You need privacy you can defend to the people whose conversations you're capturing, especially when you're talking to executives or sensitive clients.

The shape of the answer is usually software-first, hardware-optional. Granola handles your scheduled meetings. Fluent handles the conversation life between them and ties patterns together across both. If you only do scheduled meetings, Granola alone is enough. If your real value is in the relationships between the meetings, you need the broader capture layer.

Companion (not memory)

If what you want is "an AI that's always with me and we talk to each other," that's a different product category. Friend is honest about being a companion. Pi (from Inflection) is conversational AI without capture. Replika is companion-shaped with longer history.

None of these are memory or productivity tools. Don't confuse the categories. You'll be disappointed by either side.

The questions nobody is asking

Most comparison articles stop at the feature matrix. Five questions worth asking before you commit:

1. What's the company's exit pattern?

Three of the five hardware pendants on the list above have been acquired in the past twelve months. Limitless went to Meta. Bee went to Amazon. Rewind quietly folded into Limitless before Limitless folded into Meta. The pattern is real. AI capture is consolidating, and your conversation data is the asset.

Before you buy, ask: is this company's business model "build a brand and exit," or is it "build a sustainable product"? The former is fine if you're early-adopter-comfortable. The latter is what you want if you plan to wear the device every day for years.

2. What's the data export story?

Every product on this list claims to let you export your data. The real question is what happens when the company sunsets, gets acquired, or pivots. Will the export endpoint still work in 90 days? Six months? Two years?

The honest answer: open-source self-hosted (Omi) is the only option where the answer is "as long as you keep the code." Everything else is a promise.

3. What's the encryption posture?

There's a meaningful difference between:

  • Cloud storage: the company stores your data and could, in principle, read it.
  • End-to-end encryption with company-held keys: encrypted, but the company can decrypt if compelled to.
  • End-to-end encryption with user-held keys: the company literally cannot read your data, even if subpoenaed.
  • Local-only: the data never leaves your device.

Marketing copy often blurs these. Read the security page carefully, and ask the company directly if it isn't clear.

4. Where does my conversation life actually happen?

Most people who buy a pendant overestimate how much value they'll get from ambient capture and underestimate how much they'd get from just better phone-call and meeting capture. If 80% of your important conversations are scheduled calls, you don't need a pendant. You need software that handles calls and meetings well.

Be honest about your conversation surface area before buying hardware.

5. What am I actually trying to remember?

Two distinct goals get conflated as "AI memory":

  • Transcripts and action items. What was decided, what's the to-do, send me the summary.
  • Patterns and relationships. Who keeps coming up, what they said three months ago, how something has changed over time.

These call for very different products. Otter, Granola, and Plaud are excellent at the first. Fluent is built for the second. Personal.ai is somewhere in between. Pick based on which problem you're actually trying to solve.

Decision tree

If we had to compress all of the above into a flowchart:

  1. Do you need ambient capture (in-person, casual conversations) or just meetings? Just meetings → Granola (Mac) or Otter (B2B). Stop here. Ambient → continue.

  2. Hardware or software-only? Hardware → continue. Software-only and you already own a Limitless or Omi pendant → Fluent. Software-only and you don't own a pendant → wait for Fluent's pendant-free mode (on the roadmap) or use Personal.ai for AI-stack-style memory.

  3. What's your privacy priority? Maximum (self-hosted, audio never touches a cloud you don't own) → Omi. Standard (cloud-stored, encrypted at rest, finished hardware) → Plaud or Fluent. Cheap and curious → Bee.

  4. What's your acquisition risk tolerance? Want to defend the brand if challenged → Omi (open-source) or Fluent (independent). Don't care, just want polish → Plaud.

Where Fluent fits

Fluent isn't a pendant. It's the iOS intelligence layer for the pendant you wear. Today, that means Limitless and Omi. More integrations are on the roadmap.

It turns whatever your pendant captures into full transcripts, meeting summaries, person profiles built from cited evidence, daily insights, and goal tracking. Two features distinguish it from the rest of the category. When someone in a conversation mentions wanting to set up time, Fluent handles the calendar follow-through automatically. And there's a user-facing MCP server, so you can plug your conversation history into Claude or ChatGPT and query it directly. Almost nothing else in the consumer space does that.

Encrypted at rest. Per-user encryption is on the roadmap.

Fluent isn't the right answer if you mostly do scheduled video meetings on a Mac (use Granola). It isn't the right answer if you want a hardware-only solution (use Omi). It isn't the right answer if you want a chat companion (use Friend or Pi). And if you don't own a Limitless or Omi pendant, Fluent isn't the right tool yet. Pair it with one first, or wait for pendant-free mode.

It's the right answer if you've been capturing conversations and have nothing useful to do with the resulting transcripts. If you want to know what someone keeps mentioning, what changed since last quarter, what a client said three months ago that you forgot to follow up on. The patterns, not just the transcripts.

When you'll need to revisit this list

Probably in twelve months. The category is consolidating, the business models are shifting, and at least one or two products on this list will be acquired, sunset, or repositioned by then.

The principle that survives the consolidation: prefer the option whose business model isn't build a pendant brand and exit. Open-source hardware (Omi) is one such path. Software that works with whatever pendant you've already bought (Fluent) is another. The rest are bets that the next acquirer will treat your data better than the last one did.