1 / 7
Tom writes to Acme
Tom, a Forager, emails Lisa at Acme about the rollout and BCCs the beehave address. That's all he ever does.
Before anything else
First, our words
You come from the world of sales, I come from building systems. You've already drawn this idea in your own words, with your six contexts around the customer, and it fits right in. Our two vocabularies only collide on one word, the one that matters most. So we start with one shared base.
When you say
"my customer"
you mean the company you sell to.
When I say
"our customer"
I mean the company that buys beehave. The companies they sell to are their clients.
That's why the bee words exist: they make it impossible to mix the two up. Our customer is the Queen. Her clients are the Fields.
Your words, our words
| You say | In beehave | Bee word |
|---|---|---|
| The company that buys beehave | Customer | the Queen |
| Their own company: people, teams, products, past deals | Home organization | the Hive |
| The companies the Queen sells to (her "customers") | Client | a Field |
| One account, one prospect | An account | a Flower |
| Everyone at the customer | Internal users | Workers |
| Everyone who works with clients | Client-facing users | Field bees |
| You: AE, account manager | Account executive | a Forager |
| The SDR | Sales development rep | a Scout |
| The contact on the client side | External contact | a Visitor |
| The colleague who never updates the CRM | no plain term: buyer talk only, with a wink. Never on a screen employees see | a Drone |
| The new hire | New team member | Brood |
| Emails, calls, notes, files | Raw information | Nectar |
| What the team knows | Knowledge (claims) | Pollen |
| Closed-won, revenue | Contract value of won deals | Honey |
| Us: you and me, the founders, and our team | The service | the Beekeepers |
One rule that never bends: Honey is always revenue. Never data, never answers. And the beekeepers, that's us, are the kind who never take the honey: the hive, the bees and the data are all hers.
Words you'll hear from me
Tap one to open it.
Systemthe whole thing
All of beehave, as one thing. One copy runs for each customer, on that customer's own server, and we keep every copy on the same version.
Subsystema big part with one job
A part so big it could almost be its own product: getting data in, getting answers out, the AI researcher, running the servers. Right now there are four, plus a health check. Not final.
Modulea smaller part inside
A part inside a subsystem. It owns its own data and has exactly one door. You can fix, improve or replace one without touching the others. That's what lets a small team build something big.
Originalthe untouched thing
The email or file exactly as it arrived, byte for byte. Kept forever and never changed, so everything else can always be rebuilt from it.
Evidencethe label on everything
Who brought it in, who said it, when, and where it came from. Every piece of data carries this card. Like a CRM activity log, except nobody has to type it.
Claimone thing someone said
"Anna works at Acme," says her email signature, March 2026. beehave never stores "the truth", only claims with their evidence. Two sources can disagree, and we show both.
Re-rungetting smarter for free
Running the stored originals through improved logic again. The memory gets better without anybody sending anything new, and old results stay readable next to the new ones.
Webhookan inbox for machines
An address another tool can send data to automatically, without anyone clicking. That's how Zapier and its cousins talk to us, and our very first way in.
Classic querya precise question
A normal, exact database question with an answer you can check. No AI guessing involved: ask it twice and you get the same answer, with its proof.
Model, LLMthe AI words
A model is any trained AI. An LLM is the kind that writes text, like ChatGPT. Useful, but it sometimes makes things up. That fact shapes half of part two.
Pipelinenot your sales pipeline
Careful, another collision. For you it's deals by stage. For me it's the line every email and file travels through: in the door, ripened, checked, stored. "The ingestion pipeline" means getting data in.
Graphthe web of who knows whom
Not a chart. A web of dots and lines: people, companies and deals are the dots; "works at", "knows" or "decides for" are the lines. It answers "who do we know at Acme?" in one step.
Where your map fits
The Forager's view of one Flower
Your first sketch put the client in the middle with six contexts around it. It was drawn from sales, and that's fine: it's one view.
The idea we agreed on at our last meeting is bigger: one hive that keeps everything everybody learns, for everyone in the company. The hive stores it all once. A lens is a set of questions you ask it. Your six contexts are one lens: the Forager looking at one Flower, one client.
Other bees get other lenses, for example:
- The Brood (new hires): what do I need to know to start?
- Scouts: who should we talk to next?
- The Queen, later: the leadership overview, her royal jelly.
Vendor & Solution
Business & Industry
Operational
One Flowerone client
Value
Decision
People
The real job
What does a rep ask the hive on day one?
Two of your questions are exactly right:
"Who matters, and what drives their decisions?"
"How will the organization decide and commit?"
The day-one questions decide what we build first. We have to get them right, and that's where your world comes in.
The name
beehave, for now
beehave is the code name: behave plus beehive. Simple and a bit cheeky. People should behave a little more anyway.
The brand name isn't decided. The trick behind all of them is the same one the code name uses: swap be for bee.
Beememberremember, and "be a member"
Beequeathpass it on to the next generation
Beecauseevery answer comes with its reason
Beelongnew people belong from day one
Wagglethe dance that shows where the good stuff is
Beeknownwhat your people know stays known, and you get known for it
PART ONE
The vision
What we'll tell the people who buy it.
The problem
Maria leaves in March
She knew who really decides at her three biggest accounts, what the works council blocks, and why last year's pilot died. What's left after she goes: a CRM record and an inbox nobody will read.
A CRM keeps the deal. A wiki keeps documents. Nothing keeps what was learned.
You've lived this from both sides: inheriting a territory, and handing one over.
~40%of a rep's week is actual selling. The rest is admin, research and typing data in.Salesforce, 2026
72%of reps feel overwhelmed by the number of tools they have to use.Gartner, 2024
28–40%of sales reps leave every year (AEs to SDRs).Bridge Group
35%fully trust their own CRM data.Salesforce, 2026
What beehave is
A hive that never forgets
A persistent, intelligent organizational memory: a living understanding of your own company and every company you work with. Everything your people learn stays, and gets smarter over time. So nobody starts from zero.
-
Your company is the Queen. beehave is her hive, on her own server. Her job is easy: shine, be admired, and let the bees do the work.
-
Every day the field bees fly out to the Fields, the companies she works with: Foragers to their accounts, Scouts to new prospects.
-
They come home with nectar: emails, calls, meetings, files. Nobody types a summary, and nobody sorts anything by hand.
-
The house bees ripen it and seal it in the honeycomb, each cell marked with who said what, when, and where.
-
What's stored is pollen: who really decides, why the pilot failed, who knows whom. New hires, the Brood, are fed it from day one.
-
And with it, the Foragers win honey: revenue. When a bee flies on, what she knew stays in the hive.
Why buyers want it
Six reasons to say yes
Nothing walks out the door
When someone leaves, what they knew stays, and their accounts are ready for the next person.
Nobody has to write anything
Knowledge comes in while people work: email, calendar, chat, calls, files, links. No logging, no "please update the CRM".
New hires start with what the team knows
Not with homework. Ramp time is the expensive part of every hire, and the hive cuts it from day one.
Every answer shows its proof
Who said it, when, and a click to the original. beehave never claims to know the truth.
It sits under the tools you already use
Not one more tool to type into. It collects what those tools already hold.
The data stays home
On the customer's own server. The AI runs on our own hardware, never through a cloud AI service.
How we'll talk to them
Long live the Queen
Welcome, Your Majesty.
Your hive is ready, and your bees are waiting.
From now on your job is easy: shine, be admired, collect your honey, the revenue. We'll handle the rest.
Where shall we begin? Connect your email, and the house bees get to work.
The voice in one line: royal treatment and bee humor on the surface, serious work underneath. We joke about the hive, never about the data.
- You rule. We buzz.
- Long live the Queen. And everything she knows.
- Don't beelieve us. Check the evidence.
- The drones never update the CRM. With beehave, they don't have to.
PART TWO
Inside the hive
The decisions I made as product owner and designer, and why.
Internal only · between us, not for buyers
System, subsystems, modules
How the hive is built
One system, beehave. Right now: four big subsystems, plus a health check. Inside them, small modules with one job each. Tap any of them.
Current status, not final: parts will be added, split or merged as we learn
Running every hiveMaster · the beekeeper's tools
Our side, above all the customer servers. Creates each new customer's server, configures it, keeps every server on the same version, and pulls encrypted backups. It never touches customer data.
Why: many customers, one package, one version. Updates go out to everyone at once.
One customer's hive (her own server)
1 · Getting data in
The entrance & Receiverscapture
Takes anything in, stores the untouched original with its evidence card, and says "got it". Nothing else: it never opens, reads or understands anything.
Why: "got it" must mean safe. The front door can never be slow, clever or broken.
Guardsaccess
Check every knock: who is this, and may they do this? Nobody gets a pass: not outsiders, not insiders, not even us.
Why: zero trust. "Internal" is a label on the evidence, not a key.
The to-do listqueue
A simple list of jobs between the steps, in its own database. If a later step breaks, nothing is lost: the job just waits.
Why: everything after the front door is allowed to be slow or down for a while.
Ripeningprocessing
Reads the file, splits an email into its parts and finds who, when and what. Six steps, two fixed hand-over formats.
Why: rules first, an AI model only where it has to understand. Any step can be swapped without touching the others.
Cleanersvalidation
Check every record against fixed rules, twice. Broken data goes back with reasons, and after a few tries to quarantine. Uncertain data passes, marked as uncertain.
Why: nothing is ever deleted, and nothing broken slips into the comb.
Honeycombstorage
Three levels: the record says what, the text says it all, the original proves it. From any claim, one click to the exact passage it came from.
Why: re-reading a file is the expensive part, and it never has to happen twice.
Matchingsame or not?
Spots that "DB" and "Deutsche Bahn", or two Anna Webers, are probably the same. It proposes, a person confirms when it's unclear, and every merge can be undone.
Why: a wrong merge silently poisons a graph. So it's a claim too, with evidence.
Resolverwhat people see
Turns all the separate statements and confirmed matches into the claims people actually see, with time periods: "worked at Acme 2015 to 2019, and again since 2022".
Why: people need one clear answer, with every source still behind it.
Enrichmentwhere the machine learning lives
Most of the machine learning lives here: ranking and re-ranking evidence into confidence, and embeddings (vectors) that find similar meaning even when the words differ, like "DB" and "Deutsche Bahn".
Why: machine learning helps sort and find. It never decides what's true.
2 · Getting answers out
Ask the hiveclassic queries
Answers questions from the claims with classic queries. Every answer can be checked without any AI. Every question runs as the person asking: you only see what you may see.
Why: the same question from two people is two different questions.
3 · The researcher
Waggle dancethe AI layer
Later: a powerful AI that understands the whole database and digs deeper in the customer's own data. Only through retrieval's door, always as the person asking. Its answers are guidance.
Why later: see "Why the AI chat comes last".
+ Health
Heartbeatsis everything alive?
Every part says "I'm alive, here are my numbers" every 30 seconds. Green, yellow or red per module, and every item can be followed through every step.
Why: we know something is broken before a customer does.
On top · later
Once the hive holds what everyone learned, new features are mostly new questions to it. Three that come almost for free:
Onboardingthe Nurse bees feed the Brood
A new hire gets their own view on day one: their accounts, who matters there, what was promised and what went wrong before. Built from what the team already knows.
Why: ramp time is the most expensive part of every hire.
Teaching & trainingthe old bees feed the young
Courses and playbooks made from real deals: how this kind of client decides, which objections came up, what won and what lost. Every lesson points to its evidence.
Why: the best lessons are already in the inbox. Nobody has to write them up.
Briefingsbefore every meeting
One page on a Flower before a call: who's in the room, what they care about, open promises and the last contact. What every Forager wishes someone had written.
Why: prep goes from an hour of digging to a minute of reading.
Field bees are people. House bees (Receivers, Guards, Cleaners, and later Nurse bees and the retinue) are the machinery.
Follow one email
From inbox to honeycomb
2 / 7
The Guards check the knock
Known sender, allowed to send. The sender's domain says this belongs to the Acme area. The way in decides the area.
3 / 7
Stored untouched, "got it"
The original goes on the shelf with its evidence card. Only then does beehave say "got it".
received2026-03-12 14:02brought in byTom Becker (internal)channelemail, BCCareaAcmesaid byTom Becker
4 / 7
Ripening
The house bees split the email: Tom's new text, his signature, the quoted reply from Lisa. Rules find names, dates and amounts. A model reads the meaning.
5 / 7
The Cleaners check
Does the record follow the rules? Yes. It goes into the comb. If it didn't, it would go back with reasons, never quietly disappear.
6 / 7
Pollen in the comb
New claims, each with its evidence: Lisa Weber is head of HR at Acme. Tom knows Lisa. The second one is the gold: the link between our side and theirs.
7 / 7
Six months later
Tom has flown on. His successor asks the hive: who do we know at Acme?
Lisa Weber, head of HR. Known by Tom Becker: 14 emails, 3 meetings, last March 2026.
Decision
Why I'm obsessed with data
beehave is a data problem, not an interface problem.
The one mistake we can't undo
Losing data. If something isn't captured, or the original isn't kept, it's gone. Everything else can be fixed later. So the front door is sacred: "got it" means safe.
The value lives in the claims
Turning messy text into the right claims and links is where the product's worth is. A wrong graph can't be saved by a nice screen.
It gets better for free
Because every original is kept, better logic can run over all of it again. The memory improves while nobody works on it.
Any data, any time, any format, any way
The hive works with anything: mail, calls, files, links. Nobody should have to think about what goes in, or how.
Decision
Why the screens come last
CRMs fail because they're messy: you have to know where to look before you can find anything.
And they fail on a wrong assumption: that busy people are smart about it and willing to do extra work. Most aren't. Not because they're bad, because they're human, and a little lazy. So adding knowledge has to be so simple that anyone can do it without thinking. Better still: without doing anything at all.
People want one of two things. A simple answer, or a trail to follow.
If a plain question to the graph gives an answer a person understands, even a rough one with words missing, we've made it. The screen then only has to show the answer and its proof. That part is easy, and it stays simple.
But when the screens do come, it takes balls: the balls to dare something completely new. Bee biology is very clear on this. The queen gathers once and stores it for years: that's our data layer, patient and safe. The drone gets exactly one flight, gives it everything he has, and it literally costs him his manhood.
The future interface is the drone's flight: a completely new way of working, not another CRM screen with a bee sticker on it. That needs balls. And balls need testes. Lots of testes, on real reps, before anything ships.
Unlike the drone, we plan to survive it.
The answer"Lisa Weber, head of HR at Acme."
The trail"Says who? When? Show me the email."
a sketch, not a design
Who do we know at Acme?
Lisa Weber, head of HR. Known by Tom Becker.
14 emails3 meetingslast: March 2026
› show the originals
In a CRM7 clicks, 3 tabs, 1 guess
In the hive1 question, 1 answer, its proof
Decision
Why the AI chat comes last
An LLM always makes things up sometimes. That can't be avoided, so it's never the source of an answer.
- Classic queries first. Every answer can be checked without AI.
- AI on top, later, only once the classic answers are good.
- Even then, only through retrieval's door, as the person asking, and every answer points back to its proof.
The common mistake is adding the AI too early, because it's convenient. We don't.
68%of a language model's labels flipped when fake "authority" text was hidden in a document.study, 2026
33%still flipped when the model could only pick from the allowed labels.same study
So the risky jobs inside the pipeline go to small models that can't write text at all, and every model is named by its job: vision reader, block labeler, meaning extractor, note reader, match helper.
Decision
Boring on purpose
Keep it simple. Every choice here is the least clever one that works.
One server per customer
Her data lives there and nowhere else. No customer ever shares a database with another.
One database
Postgres remembers, memory thinks: one database is the record, and a graph in memory answers "who knows whom".
AI on our own ground
On the customer's server or our own hardware, which keeps nothing. Never a cloud AI service.
Few dependencies
Boring technology, our own thin clients, no frontend frameworks. Less to break, less to attack.
One door per module
Each module owns its data and can be rebuilt alone. A future buyer and their auditors will read this code.
Nothing is rewritten
Originals forever, every version kept, every merge undoable. You can always see what changed, and why.
Decision
Getting data in: don't wait
Start with what works today. Native connections come later, one by one.
A Gmail "connect" button needs a Google security review: 6 to 12 weeks, renewed every year. We don't wait for that. Gmail works from day one through a forwarding rule or Zapier.
Day one
- One webhook for Zapier, Make, n8n and Power Automate: a company links its tools in minutes.
- Forwarding and BCC to the beehave address.
- Calendar invites to the same address.
- Uploads and shared links (Dropbox, Google Drive, OneDrive).
Later, step by step
- Connect buttons: Microsoft 365, Slack, HubSpot, Salesforce, Dropbox. Gmail once Google's review is through.
Insider
Features don't pay. Clients do.
Everyone builds features. Building got easy; getting clients got harder. I'm not building features. I'm building an app that works and shows real benefits fast, so we get paying clients.
Where we are
What's real, and what I need from you
No product code yet. What exists: 93 written decisions, a vocabulary that is defined rather than sketched, the whole pipeline with its hand-overs, and research checked against sources.
The next step is not another document. It's one real customer's mail run through the whole pipeline, and one question answered end to end.
Why does all of this look so easy? Because it's over 14 years of solving exactly these problems, folded into a few bees. Nothing here is new on its own: it's everything that already exists, rethought. Only experience makes it look this simple.
And none of it matters until a client pays for it.
Dan, this is where you come in
- The first questions. You already drafted two. What does a rep ask the hive on day one? Getting these right decides what we build first.
- Your tools. Where does your team really live: Outlook, Gmail, Salesforce, HubSpot, Gong? That's our first connection.
- Your words. How reps actually talk about handovers, lost accounts and ramp time.
- The name. Beemember, Beequeath, or stay beehave? Gut call.