AGPLv3 open coreSingle-tenant per userHosted in GermanyExport anytime

business · early access

One memory for your company. Visible, correctable, and yours.

Your team and your AI agents write into one shared vault. What one person tells one tool becomes company memory every other tool can use. Each entry records who added it. The whole vault runs on a dedicated machine for your company alone, hosted in Germany or on your own hardware, and exports in full at any time.

One machine per companyEvery entry attributedAGPLv3 open core
one company memory
a founderrecords the pricing decision from Claude
a sales leadadds the customer's constraint from ChatGPT
an agentpicks both up in the next run, attributed
one company memory, every writer on the record
01 · the problem

Company knowledge lives in personal chat histories.

Every person on your team explains the company to their own AI tools: the product, the customers, the decisions already made. That context sits in private chat histories nobody else can reach. When someone switches tools, it resets. When someone leaves, it walks out the door. And every agent you deploy starts from a blank page.

Sales
✕ starts empty
Engineering
✕ starts empty
Agent
✕ starts empty
02 · how it works

One vault. One machine. One company.

An afair Business vault is a real, dedicated machine: one instance, one database, for one legal entity. Not a workspace inside a pooled SaaS database. Your people and your agents connect through MCP, each with their own token, and write into the same shared memory. afair works out what belongs together in the background and keeps the company's current understanding available to every connected tool. The substrate is append-only: memory is added and superseded, never silently rewritten.

one shared substrate
member tokenwrites into the company vault
agent tokenreads the current state, writes back its work
every toolreads the same understanding, attributed
one machine, one database, one legal entity
03 · who wrote that

Every entry answers: who put this here?

Each member and each agent writes with its own token, and every write records which token wrote it. When an agent relays for several people, it can additionally record on whose behalf. When a tool recalls a fact, you can see who put it there, what supports it, and what changed since.

Your vault admin reviews syntheses, sources, open conflicts, and stale evidence in the Memory Mirror and corrects wrong claims there. A correction is appended as new evidence. History stays intact, which also means it stays auditable.

who asserted this
by member tokencustomer signed a two-year term
by agent tokenrenewal reminder scheduled for Q3
supersededearlier one-year note, kept in history
every claim points back to a writer
04 · what stays yours

The company memory is yours. So is the evidence.

Memory inside a model vendor's product belongs to the vendor and rewrites itself out of sight. Memory in a pooled SaaS store sits in someone else's database next to other companies. A company memory should be visible, correctable, and owned by the company. That is the whole design.

one machine, one company

A dedicated instance and database per legal entity. Never a pooled multi-tenant store shared with other companies.

hosted in Germany, or by you

Managed vaults run on dedicated machines in Germany under the GDPR. Or run the same code on your own hardware.

export everything, any time

The complete vault in one export, readable without afair software. Leaving is a feature, not a negotiation.

speaks MCP

Members and agents connect through MCP, the open protocol the major AI tools speak. No proprietary SDK between you and your memory.

AGPLv3 open core

The vault core is open source. What holds your company memory can be read, audited, and run without us.

append-only evidence trail

Corrections supersede, they never erase. What the company memory believed, and why, stays reconstructable.

05 · where it fits

One memory, every part of the company.

The same shared vault, read and written by the people and the agents already doing the work.

deal memory for a sales team

Customer constraints, objections, and commitments land in one vault as they come up. The next call prep, in whatever tool, starts from the current state of the deal.

decision log for a founder team

What was decided, when, and by whom, captured as a side effect of normal work. Six months later the reasoning is still there, with its sources.

client context for an agency

Brand rules, past feedback, and project history stay current across everyone working the account, instead of living in one person's chat history.

an agent working alongside the team

An agent connects with its own token, reads the company's current understanding, and writes back what it did. Its entries are attributed like anyone else's.

early access

Bring your company's memory home.

afair Business is in early access with a small number of companies. Tell us what your team looks like and which tools you use, and we will get back to you personally. No public pricing yet; the first companies shape it with us.

By submitting, you agree to processing of your name, email, and message as described in our privacy notice.

06 · questions

Straight answers.

Is this a multi-tenant SaaS?
No. Each company gets its own machine and its own database. There is no pooled store, no shared compute, no row of yours next to another company's.
Who can see what?
In the current version, every member token can read the whole company vault. afair Business v1 is built on attribution, not permissions: everyone sees everything, and every entry records which token added it. Finer-grained access control is on the roadmap and will be its own deliberate design, not a checkbox.
Can we self-host?
Yes. The core is open source under the AGPLv3. Self-hosted and managed vaults run the same code and the same vault format, so you can move between them.
Where is our data?
On your own machine if you self-host. On a dedicated machine hosted in Germany if we run it for you. In both cases the vault is encrypted at rest and exportable in full.
How do our AI tools and agents connect?
Through MCP. Each member and each agent gets its own token, and every write records which token it came from. Any MCP-capable client can read and write the vault.
How is this different from the memory inside a single AI product?
Built-in memory serves one vendor's product and can rewrite itself without a trail. The afair vault serves every tool the company uses, keeps an append-only evidence trail, shows which token asserted what, and leaves with you if you go.
What does it cost?
Business pricing is being worked out with the early-access companies and depends on what a dedicated company vault actually costs to run. afair Personal, one vault for one person, is 10 euro a month.