MENSARA.AI · A GOVERNED MIND, IN YOUR ACCOUNT

MENSARA · BY XANADU LABS

A Mind of your own.

Mensara is a shared memory for your company: one governed place where your team and every agent you run read the same context, work the same task list, and see only what each is allowed to see. Hosted for convenience. Owned by you: your account, your keys, and a signed receipt for every access.

The founding cohort is small on purpose. Each Mind is brought up personally.

01 · THE PROBLEM

AI can do real work for your business.

It needs three things you don't have time to build.

Context

Your edge lives in a few heads and ten years of decisions. Every AI tool you hire starts blank.

Controls

Which people, agents, and models see what? What did it read before it answered? How do you budget what it spends? Without those answers, AI stays away from real work.

Capability

The infrastructure has to be built, governed, and kept current. You already have a company to run.

02 · WHAT IT IS

One memory. Everyone works from it.

Underneath, Mensara is deliberately boring: your files, in your account, behind one gateway that decides who reads or writes them.

The machine underneath A git repository holds the record. A deterministic gateway sits in front of it and does three jobs from that one source: shared context, shared task state, and access control. The record PLAIN FILES GIT REPOSITORY The gateway DETERMINISTIC POLICY, NOT A MODEL Shared context EVERYONE READS THE SAME TRUTH Shared task state WORK TRACKED WITH HISTORY Access control ON THE RECORD, EXPLICIT
One record, one gateway, three jobs. That is the whole machine.

03 · BEFORE AND AFTER

Same agent. Different Monday.

A real exchange pattern from the Mind that runs our own company, lightly anonymized. The only difference between the two panels is what the agent can read.

Without a Mind

Draft the follow-up note to the client sponsor on the Phase 2 proposal.

Happy to help. Could you share who the sponsor is, what Phase 2 covers, the pricing you proposed, and anything relevant from your earlier calls?

You are the context. Again.

With a Mind

Draft the follow-up note to the client sponsor on the Phase 2 proposal.

Drafted. It picks up from Tuesday's call, keeps scope to the two workstreams already agreed, quotes the phase price approved on the 14th, and leaves out the staffing ask the sponsor cut last time.

read: engagement/state.md · pricing/phase-2.md · calls/2026-07-14.md · sealed 0x4e2a…9c

04 · THE GATE

Policy decides. Not a model.

Every read and write passes one gate. The gateway checks who is asking and what their capability allows, serves exactly the files policy grants, and holds everything else. Both outcomes leave a signed receipt. By the time any model is involved, the access question is already answered.

How access works You, your team, and your agents ask through the gateway. The gateway checks identity, capability, and policy, then serves granted files from the record in your account and holds everything else. Every grant and every hold is written to a signed, hash-chained receipt log. WHO ASKS You CHAT · PORTAL Your team SAME RECORD Your agents MCP · SCOPED TOKENS The gateway DETERMINISTIC POLICY 1 · IDENTITY CHECK 2 · CAPABILITY CHECK 3 · POLICY MATCH NO MODEL DECIDES HERE GRANTS ONLY YOUR ACCOUNT The record PLAIN FILES · GIT HISTORY EVERY PERMISSION EXPLICIT HELD OUT OF POLICY · The receipt log EVERY ACCESS · SIGNED · HASH-CHAINED
Access is decided by policy before any model runs. Grants and holds both leave signed receipts.

05 · NOT RETRIEVAL

A record, not a guess.

Most AI memory is retrieval: embed everything, search for what looks similar, hope the model got the right chunks. A Mind is a record. Reads return whole, governed files with their history, so an answer either stands on the record or says plainly that the record is silent. "Not on the record" is an answer you can act on. A similarity score is not.

Writes are proposals. An agent or a teammate drafts the change, a human approves it, and git keeps every version. You can diff your company's memory the way you diff code, and roll it back the same way.

06 · GETTING STARTED

Concrete from day one.

  1. Day one

    A working session, not an upload form.

    Your Mind starts as a two-hour interview. It asks what you are carrying, where the knowledge actually lives, and what you decide over and over. Then it proves itself in the same sitting: three real questions from your work, answered with receipts.

  2. The first weeks

    The move-in.

    Point it at what already exists: docs, decision threads, project folders. It structures what it reads and flags contradictions for a human to settle. Nothing gets ripped out. Your current tools keep working while the Mind becomes the place people and agents check first.

  3. Ongoing

    The upkeep.

    Knowledge rots. Your Mind flags what looks stale, drafts the update, and waits for a human yes. Every change lands as a reviewed commit, so the record stays worth trusting.

Evaluating this for a larger company? Start with one team and one workstream. The repository is clonable and inspectable from the first day, so evaluation is just usage: no migration project, no platform commitment, and it leaves with you intact if you walk away. What the first session covers

07 · RECEIPTS

Don't trust it. Check it.

Every access leaves a signed, hash-chained receipt: what was asked, what was read, what was held at the gate. And when something is not on the record, your Mind says so. "Not on the record" is an answer, not a failure. Visible imperfection beats claimed perfection.

08 · IN NUMBERS

We run on it. This is the ledger.

Xanadu Labs has run on its own Mind every working day since January 2026. These numbers are counted from that repository, not projected.

6

months of daily use

13,866

commits to the record

1,350+

tasks tracked between humans and agents

88

decisions recorded with rationale

Counted 2026-07-29 from the live repository, which also holds 8,944 files of working context. Ask us to show you around it.

09 · WHO IT IS FOR

Built for the people who carry the company in their heads.

The founder everyone routes through. The partner who is the institutional memory. The professional whose practice is their edge. If your work is knowledge and your knowledge is scattered, a Mind is for you, whether you run five hundred people or just yourself. It is not for personal life admin, and not for anyone who wants AI they never have to look at: a Mind earns autonomy rung by rung, on a track record you can read.

For agent builders

A governed MCP endpoint your agents connect to: capability-scoped tokens, whole-file reads, deterministic refusals. Plain files and git underneath, so your stack stays yours.

For founders and operators

You are the context bus: every tool, hire, and agent gets briefed out of your head. A Mind takes that job, so decisions stop queueing on your memory.

For IT and security

Explicit permissions, signed hash-chained receipts, no model in the access path, and exit by construction: the repository sits in your account, not ours.

I built this for myself first.

I ran a company from my own head for years, and I know what that weighs. Mensara is the system I built to put it down. My own company has run on it every day since. You are not buying a pitch, you are buying the thing I use.

Antony Evans, founder, Xanadu Labs.

10 · OWN IT

Yours. Provably.

Mensara runs the machinery so you do not have to. But the Mind itself, the repository, the keys, the receipts, sits in your account. Clone it any afternoon. Read every file. And if you ever leave, it leaves with you, working. We refuse to hold it for you on principle: what someone runs for you, they hold over you. Not here.

your accountyour keysleave anytime, intact

MENSARA.AI · MACHINE SURFACE · TEXT/MARKDOWN

# Mensara

Mensara is a company Mind you own. A shared memory for your company: one governed git repository your team and your agents read from and write to. Hosted by Mensara for convenience, owned by you in your account, with a signed receipt for every access.

## What it is

A git repository of plain files plus a deterministic gateway in front of it. Three jobs: shared context (people and agents work from one source of truth), shared task state (work tracked between humans and agents with full history), and access control (explicit permissions, every access logged, no language model decides at the gate).

## How access works

Every read and write passes the gateway: identity check, capability check, policy match. Grants serve whole governed files; anything out of policy is held at the gate. Both outcomes leave signed, hash-chained receipts. Details: /how-it-works/the-gateway

## Not retrieval

Reads return whole governed files with history, not similarity-ranked chunks. "Not on the record" is a first-class answer. Writes are proposals: drafted by an agent or teammate, approved by a human, committed to git.

## Getting started

Day one is a two-hour working interview that ends with three real questions answered with receipts. In the first weeks, existing docs and threads are structured in; nothing is ripped out. Ongoing, the Mind flags stale knowledge and proposes updates a human approves. Larger companies pilot with one team; the repository is clonable from day one and leaves intact.

## In numbers

Xanadu Labs runs on its own Mind. Counted 2026-07-29 from the live repository: 6 months of daily use, 13,866 commits, 8,944 files, 1,350+ tasks tracked between humans and agents, 88 recorded decisions.

## The founding cohort

Minds are brought up personally. Request a seat at /#founding-cohort with an email and a line about what you carry.

## Trust boundary

Mensara's host stores account, enrollment, billing, fleet, and coarse health records. It does not hold customer Mind content, chat history, keys, zone files, or credentials that enter a provisioned box.