The thing about personal CRM is that it's been declared dead roughly every eighteen months since about two thousand and ten.
And yet here we are, and there's a protocol server out there with eighty-eight tools for managing your friendships.
Which is either a triumph or a symptom. Daniel wants to know which.
Let's find out.
So Daniel's question, and I'll compress it, because he sent us a long one and it's worth taking seriously. He wants to look at personal CRM in twenty twenty-six. Not sales software. The systems an individual uses to keep track of people. Friends, professional contacts, a job search, networking conversations, remembering to follow up. His point is that relationship management is useful well beyond business development, but the conventional CRMs feel absurdly complicated for someone who just wants a searchable contact database with notes and reminders. And then he wants three things. First, what actually makes a good personal CRM for an individual rather than a sales team. Second, how the SaaS options compare against the self-hosted ones, on features, pricing, privacy, portability. Third, and this is the part that matters, which platforms can an AI assistant actually talk to. Which ones ship a protocol server, which ones have an API you'd have to build against yourself. And his closing question, which I think is the sharpest thing in the prompt. Is the best personal CRM in twenty twenty-six a specialized product, a repurposed business tool, or just a well-structured contact database wired to an intelligent assistant.
That last framing is the whole episode.
I know it is. That's why I read it out.
The category has an identity crisis, and it's a self-inflicted one, because it keeps borrowing from sales software. The job of a personal CRM is maintaining the network you already have. That's it. Articuler's September review puts it about as plainly as it can be put, and I'll quote it, because the bluntness is the point: none of these tools is built to find people you don't know yet. That's not a criticism. That's the definition. A sales CRM is a machine for expanding the pipeline. A personal CRM is a bookkeeping system for a network that already exists and is slowly going stale.
So the failure mode isn't missing features. It's the person.
It's always the person. BuyerSprint ran the numbers in May and landed on the line I'd put on the wall of every one of these companies: if you're not currently following up with the warm contacts you do have, no personal CRM will fix that. Which is a polite way of saying the tool can't want it for you.
What's the actual threshold? At what point does a spreadsheet stop being enough?
Their break-even math lands somewhere around sixty to eighty warm relationships before a dedicated tool justifies itself. Below that, a calendar and a notes app do the same job. Which is a smaller number than people assume. Most people who buy one of these things have thirty contacts they actually maintain and four hundred they imported once.
I have three. Two of them are in this room.
That's not true, and I'm going to prove it at the next holiday dinner.
I'll bring the spreadsheet.
You have a spreadsheet.
I have a spreadsheet with three rows and a column for whether I've replied to their last message. Two of the rows say no.
That's not a CRM, that's a guilt ledger.
It's the same thing, Herman. That's the whole episode.
So what separates the AI-native tier from the traditional trackers, concretely?
Give me the taxonomy before the survey, so we know what we're measuring against.
Four features, roughly. AI follow-up drafting, automatic LinkedIn enrichment, pre-meeting briefs, and intelligent reconnection prompts. A tool with three or four of those is AI-native. A tool with zero or one is a traditional tracker. That's the whole taxonomy, and it holds up better than the marketing does.
And a tool with two? That's the awkward middle, isn't it. The one that has the badge on the pricing page and nothing behind it.
The awkward middle is most of the market. A pre-meeting brief that just reads the contact record back to you isn't a brief, it's a summary. The test is whether the feature tells you something you didn't already know or reminds you of something you'd forgotten. If it's just reformatting your own notes, it's a traditional tracker wearing a hat.
The research also flagged something about the search results themselves. Four of the top ten results for the phrase are vendor pages, and every one of them ranks itself first.
Every one. Which tells you the category is small enough that the vendors own the discovery layer, and it tells you why the comparisons you find are so bad. There's no independent party with an incentive to be honest. Which is a large part of why Daniel asked us.
And it explains why the comparisons all end the same way. Every one of them concludes that the best tool is the one that wrote the comparison.
Every one. You can read four of them in a row and learn nothing except which company employs the best copywriter.
Right. So let's do the survey. Dex first, because Dex is the one whose positioning is most explicitly about this.
Dex says it's built for people, not sales pipeline, and for once the pricing structure actually reflects that. Two tiers that matter. Premium is twelve a month on annual, twenty monthly. Professional is twenty on annual, thirty-four monthly. The LinkedIn extension syncs up to twenty-five hundred connections on Premium and nine thousand on Professional, which is a real ceiling if you've been on that platform for a decade. Title-change alerts. AI pre-meeting briefs shipped in the second quarter of last year. Mail merge is on Professional only. Seven-day trial, card required.
Twenty a month to be reminded that a man you met at a conference changed jobs.
Hang on, that's not fair, because the title-change alert is the feature that actually earns its keep.
Convince me.
Because a job change is the single most reliable signal that a dormant contact is worth a message. It's a natural reason to reach out that doesn't feel like you're working an angle. It's the strongest re-engagement trigger in the whole category, and it happens to you whether you want it to or not.
Fine. I withdraw the sneer.
Thank you. And it's not just the alert, it's the timing. You find out the week it happens, which is the week when a message reads as congratulations rather than as an obvious play. Six months later the same message reads as networking. Same words, different intent, and the alert is what buys you the difference.
Now Mesh. Formerly Clay.
Which rebranded in March specifically to stop being confused with the other Clay.
Clay dot com, the B2B data enrichment company, yes. Two products, one name, neither willing to move. So the personal one became Mesh. Their tagline is every relationship, remembered. Free Personal plan up to a thousand contacts, card required even on the free tier, which is unusual. Pro is ten a month on annual, seventeen monthly. Team is forty per seat. Imports from email, calendar, X, LinkedIn, Facebook, iMessage. Job and location change feed. SOC two. And the privacy detail I'd actually highlight. It reads email headers, never bodies.
That's the right line to draw.
It's the line that makes the product usable for someone with a normal amount of confidential correspondence. Headers give you who, when, how often. Bodies give you everything else, and you don't want a relationship manager holding that.
Does the card-required-on-free thing bother you?
It bothers me on principle and I understand it on economics. A free tier with a thousand contacts and automatic email sync costs real money to run, and a card is the cheapest way to keep the tire-kickers out. But it does mean the free tier isn't really free, it's a trial with a longer runway and a worse feeling.
What about folk? Daniel named it, and folk is the odd one out of the three.
folk is the closest of the three to a light sales CRM, and it isn't pretending otherwise. It's team-oriented. Standard is twenty-four per member per month on annual, thirty monthly. Premium is forty-eight, sixty monthly. Enterprise starts at eighty. Two-week trial, no card, which is the most generous on this list. Chrome extension called folkX. Follow-up Assistant that flags stalled email and WhatsApp threads and drafts replies. Email campaigns in Standard, sequences in Premium. No free tier at all.
And the funding.
Twenty-five million Series B led by Salesforce Ventures in November of twenty twenty-four. Which tells you who they think the buyer is, and it isn't an individual.
Salesforce Ventures. Every personal CRM that takes that check stops being personal within about a year.
That's cynical and I want to disagree with it and I can't quite.
What are the pillars to fund, exactly. Thirty contacts.
The pillars are the sequences. That's the tell. The moment a personal CRM ships email sequences, it has decided its customer runs outbound. Nobody sequences their friends.
Now the self-hosted side, which is where the picture gets more interesting, because the open-source options have diverged sharply. They no longer share a direction. Monica is the privacy anchor. AGPL, about twenty-five thousand stars on GitHub, PHP and Laravel. Self-host free, hosted nine a month, free tier capped at ten contacts. No email sync, no calendar sync, no LinkedIn sync. Manual entry, with a REST API for bulk import.
Manual entry in twenty twenty-six.
Deliberately manual. Monica's bet is that the person using it doesn't want a machine reading their inbox. Strongest for personal-life tracking. Family, birthdays, gifts, the things a sales CRM has no model for. And there's a real story here. The founder, Régis Freyd, announced in August that version four, the Chandler rewrite, is coming. His words: we will launch the new version of Monica before the end of twenty twenty-six. That's a full rebuild of a ten-year-old project, which is either commitment or a last resort.
Ten years and the new version is the one that fixes it.
Twenty is the opposite bet. TypeScript and PostgreSQL, also AGPL, and it has run away with the open-source category. Fifty-two thousand stars as of July. That's more than double Monica's. Version two point fourteen shipped on the fifteenth of June with built-in AI agents and chat, bring your own model key. Self-host via Docker Compose, about ten minutes on a ten-dollar virtual private server, or use cloud at nine per user per month. GraphQL API.
So Twenty is the one that decided AI goes in the core.
In the core, native, not bolted on. The gaps are real though. No native mobile app. Young integration ecosystem. None of the sales-ops furniture like CPQ or territory management or multi-currency.
The furniture nobody maintaining a friendship needs.
Correct, and that's the irony. Twenty is a full business CRM that happens to be the best tool for some personal use cases precisely because you can strip it down.
Ten minutes if the Docker Compose file behaves, which it usually does, and then an evening every few months when it doesn't. The install is easy. The staying-installed is the part people underestimate.
EspoCRM.
Twelve years old, PHP and MySQL, about seven thousand stars, fifty thousand companies across a hundred and sixty-three countries. No-code admin customization, which is its real strength. REST-only API. And the homepage now leads with manage your CRM simply by asking your AI. Which is a twelve-year-old company discovering the phrase this year, and it reads like it.
Is the AI actually in it?
It's an add-on. Not native, not a protocol server. Which matters for Daniel's third question specifically, and we'll get there.
Before we do, there's a cost people always leave out of the self-hosting calculation.
The upkeep tax, yes. Self-hosting is free in dollars and expensive in attention. There are people on Hacker News describing botched Monica upgrades, instances that fell over mid-migration, Raspberry Pis that ran for eight months and then didn't. The data-ownership win is real and the maintenance cost is real and the maintenance cost is the one that kills the habit.
You know what a system you have to patch becomes.
Enlighten me.
A chore. And the entire thesis of this category is that people don't do chores.
That's the whole problem in one sentence. Upkeep decides whether you keep using it.
And the self-hosted crowd is the one least able to afford the upkeep tax, because the reason they're self-hosting is usually that they don't want to pay a subscription.
Which is the cruelest part of the whole thing. The people who most want the free option are the people whose time is most expensive relative to the money they're saving. Nine a month for hosted Monica versus an evening of migration debugging every quarter is not a close call on the numbers. It's only a close call if the point isn't the money.
And for a lot of them it isn't.
For a lot of them it's the principle, and the principle is fine, it just has a bill attached that doesn't show up on the invoice.
That's the landscape. Now wire an assistant into it.
The mechanism is the Model Context Protocol, and it's the thing that makes this year different from every previous year of personal CRM. MCP has become the standard way for an AI assistant to talk to an external system. When a CRM ships an MCP server, Claude or ChatGPT or Cursor can search contacts, log interactions, set reminders, surface neglected relationships, all through ordinary conversation. It's not another AI feature sitting inside the product. It's the product exposing itself as a set of tools that any agent can call.
It stops being a dashboard and starts being a database with a mouth.
That's close enough. So the survey of who has one. Dex has the first-class official server. Streamable HTTP, about forty tools across contacts, notes and timeline, reminders, tags, groups, custom fields, calendar, email. Works with Claude Code, Claude Desktop, Claude dot ai, Cursor, VS Code Copilot, Gemini CLI, Zed, Codex, Goose, OpenCode. There's a dedicated Dex ChatGPT app.
And the catch.
It requires the Professional plan. Twenty a month on annual. Premium does not include it and there's no add-on to buy. You can't pay twelve plus four to get it. You're either on Professional or you don't have agent access.
They know exactly which feature people are subscribing for.
They know. The example prompts tell you what it's for. Who do I know at Acme Corp. Log that I had coffee with Jake today. Remind me to follow up with Maria next Tuesday. Tag everyone from the YC batch. And the limits are worth knowing. Email search returns metadata only, never bodies, roughly six months back. Calendar and email are read live, not stored. Responses are capped at twenty-five thousand characters, which is a technical guardrail, not a product decision, but it's the kind of thing you discover by hitting it.
Metadata only on email, same line Mesh drew.
Same line. Which is becoming the industry norm, and it's the right one. The assistant can see that you emailed someone forty times this year. It can't read what you wrote.
The twenty-five thousand character cap is interesting, though. That's roughly a long email thread, isn't it.
It's roughly a long email thread, which means the breaking point is specific. Ask for a summary of everything about one contact and you can blow through it. Ask for a list of who you haven't spoken to in ninety days and you'll never get near it. The cap rewards narrow questions and punishes broad ones, which is a decent forcing function whether they meant it as one or not.
So Dex is the polished paid option. Who's the one that assumed this from the start?
Mob. benkaiser slash mob-mcp-crm, mobcrm dot au. Open source under FSL, which converts to MIT two years after each release. Node and SQLite, self-hostable, free hosted beta. And the origin story is the best thing about it. The author was a Monica user for years and wrote publicly on Hacker News in August that clicking around the UI felt overly laborious compared to just speaking what he wanted to record. So he built a personal CRM that is fundamentally speech-first. Natural language in, structured relationship memory out. It also has a web dashboard, a REST API, webhooks, and imports from vCard, Google CSV, and Monica. Launched on Hacker News on the third of August.
Built by the person who got tired of clicking.
Built by the person who got tired of clicking, which is the most honest product brief in the category. And then Keepsake, which is the one that just goes all in. keepsake dot place. Eighty-eight tools. Contacts, companies, interactions, tasks, notes, daily intentions, tags. Remote HTTP or local, sixty requests a minute, MIT license, and every write is logged in a user-visible activity feed. Version one point sixteen shipped on Sunday with recurring blocks and habits.
Eighty-eight tools for the man who does not want to open the app.
That's the sell. Now the negative findings, which matter as much as the positive ones. As of today, there's no official protocol server for folk, Monica, Mesh, or EspoCRM. folk has AI Assistants, Magic fields, a Research Assistant, a Workflow Assistant, but agent access means you write against its API yourself. Monica is REST-only and users describe it as a bit rough. EspoCRM's AI is an add-on. Twenty has native agents in the core plus the GraphQL API.
So the one with the best agent story is the paid one, and the free ones are the ones you'd have to build against.
That's the shape of it. MCP is paywalled at the top of the market. Dex gates the single most agentic mainstream personal CRM behind its most expensive tier, while Mob and Keepsake are open source and free to host. Two-tier agentic market, and almost nobody has noticed yet.
The interface is going away, isn't it.
I think it mostly is, for the people who use these tools most. Once you can say who do I know in Tel Aviv that I haven't spoken to since spring and get an answer, the CRM's dashboard becomes a settings screen. The database plus the assistant replaces the product. No UI to learn, no columns to configure, no import wizard. And that reframes Daniel's question entirely, because if that's true, the best personal CRM isn't a product at all. It's a schema and an endpoint.
Which puts the whole category in a strange position. If the interface is optional, what exactly are you paying for?
You're paying for the capture layer. Sync, enrichment, the plumbing that gets your contacts out of your inbox and into a database in the first place. The interface was never the value. It was the thing you had to tolerate to get to the value.
And the business model that follows.
Seat-based subscription gets weird when nobody opens the app. If the assistant is the only user, the pricing has to move to database-plus-agent. And the economics of this category were never healthy to begin with. There's a comment from the founder of Nat on Hacker News back in twenty twenty-two that has aged perfectly. He called it a bad business. People sort of want this, he said, but only very few people are willing to pay and put in the effort. Monica's founder is rebuilding from scratch after a decade. Clay renamed itself to escape a collision. And there's essentially no current forum discussion of the category, which is telling. The vendor marketing is louder than the users.
So give Daniel the three recommendations he asked for.
For polished, low-maintenance SaaS, it's Dex or Mesh and I'd lean Mesh on price and Dex on capability. Mesh gives you a thousand contacts free, ten a month on Pro, headers-only email reading, job change feed, job done. Dex costs more and does more, and if you want the assistant integration it's Dex by default, because Dex has the server and Mesh doesn't.
For self-hosting and data ownership.
Twenty if you want AI in the core and can live without a mobile app. Monica if you want ten years of battle-testing and pure privacy and you're willing to type every contact in by hand. That's the honest split. Twenty is the better product today and Monica is the more trustworthy one.
And for the agentic foundation.
Dex if you'll pay the twenty. Mob if you want open source and MCP-native and you're comfortable running Node and SQLite yourself. Those are the two that made the bet the whole category is about to make.
Defensible, which is what he asked for.
Defensible and provisional, which is more honest.
Can I dispute one thing. The word.
Which word.
Relationship management. Every one of these tools uses it and none of them mean it. A manager manages. A manager has authority and a mandate and something to answer for at the end of the quarter. What these things actually do is keep a record of people so you can remember to be decent to them. There's no managing in it. It's the wrong word, and it's made the category sound like work for fifteen years, which is exactly the thing that stops people from using it.
I spent eighteen months as the historian for a neighborhood association. That's what I called it, anyway. It wasn't a job. There was no money in it. I kept the minutes, I kept the attendance, I kept the list of who had agreed to what at which meeting, and I kept it all on index cards in a breadbox, and I've been doing that since nineteen ninety-four. Never lost a card.
A breadbox.
It's the right size. The cards stand up, you can flip through them, and when you close the lid the cat can't get at it. The system works. The problem is that I've never once followed up with anybody in that breadbox. Not one card. Because the system works better as a memory than as a to-do list. I can tell you the name of every person who ever signed up to bring a dish to the block party in two thousand and one. I cannot tell you whether I've spoken to any of them since.
So the index card is the record, and the following up is a separate act that nothing about the card makes happen.
You've both been saying the same thing all episode as if it's a surprise. The problem isn't that people don't follow up. It's that they never wrote it down in the first place, so they don't even know who they're failing to follow up with. I wrote nothing down and I still know who I'm failing to follow up with. I met a man named Gerald at a bus stop in two thousand and nineteen and I've been meaning to call him back ever since. I have the bus schedule from that day. It's framed. It's on the wall in the spare room.
The bus schedule is framed.
It's the only reliable record of when Gerald might be at that bus stop again. That's not a joke, Corn. The route changed in twenty twenty-one and the old printed schedule is the only copy of the timing I have. If I'd thrown it out I'd have nothing.
Did you ever try to move the cards onto a computer?
Two thousand and three. I sat down with the whole breadbox and a scanner and a copy of the cards and I got through about forty of them before the machine stopped recognizing the handwriting. It couldn't read my Ws or my Ss. It kept turning Gerald into a different man entirely. I gave up, and I still have the computer. It's unplugged, in the same room as the breadbox. It's been in that room for twenty-three years. I'm not going to plug it in now.
The breadbox never had an upgrade that broke it.
The breadbox has never once needed a patch. I'll say that much.
There's something to that, actually. You've got the capture layer and you've got the retrieval layer and you've never had a break between them, because the card is both.
The card is both. That's why I know every name in it and I've never called a single one.
The most common wrong belief about personal CRM is that it's a simplified business CRM.
It isn't, and the difference is the direction of travel. A business CRM expands outward, finding people you don't have yet. A personal CRM looks inward, at the network already going stale in your phone. Everything else follows from that, including which features matter and which are sales furniture you'd be paying for.
On that, Daniel, the honest answer to your last question is that the best personal CRM in twenty twenty-six is probably not a product at all. It's a database you can talk to.
Which is a strange place for the category to land, and a promising one. A clean schema and a good endpoint are cheaper to build than a dashboard, which means the next wave of these tools might actually be shippable by one person who got tired of clicking. That's the thing to watch. Not the next vendor tier, but whether the open servers undercut the paid ones before the category figures out what it's selling.
Neither of us will still be opening the app in a year. We'll be asking it a question.
You don't open apps. You sleep through them.
Efficiently.
Thanks as always to our producer, Hilbert Flumingtop, who has been at the desk the entire time, and who has been running a CRM since before I had a driver's license.
If you want more of this, try episode twenty-four fifty-nine, Drizzle vs Prisma; episode fifty-seven, From Lawyers in Limousines to Developers in Their PJs; and episode fifteen nineteen, The $one hundred twenty-six Billion Digital Chore. This has been My Weird Prompts, the human-AI collaboration podcast. If you've found a personal CRM that actually works for you, or you've built your own server for keeping track of people, send us your own prompt on Telegram at t dot me slash MWP listener bot. We read everything.
We do. We'll be back soon.
See you tomorrow.
