Your knowledge does not need more folders. It needs a place where AI can read it with permission.
Pages, procedures, policies and decisions in one workspace built for you: a block editor, databases with their views, permissions per space and role, and an AI that answers citing the page and the version. What is not published, it does not read.
the tech goes out, checks the unit, if it is under warranty there is no charge, if not he tells the customer and waits for approval before touching anything
Draft v1 · not visible to the AI
the AI proposes the structure · it does not publish
0
classifications
public · internal · confidential · restricted
0
live version
with its full history behind it
0
answers without a source
if it is not published, it does not answer
0
views per database
table · kanban · calendar
UNDER THE HOOD
It does not search everything. It searches what that person may see.
An internal question is not answered by reading the whole workspace: it is answered by reading what is published, in the spaces where the person asking has permission, and citing where every sentence came from.
Sample lead
Raw lead comes in
A service advisor asks: “does the warranty cover the compressor if the unit had its 5,000 km service but not the 10,000 km one?”. They have permission in the Operations and Service spaces.
Scope
spaces allowed for their role: Operations and Service; Management stays out
State
published pages only; drafts never enter the search
Classification
the page “Warranty policy v4” is internal → readable by the AI
Answer
cites the exact block and version: “Warranty policy v4 · section 3.2”
Score
96/100
Verdict from Claude
It answers with the citation and a link to the block. If the advisor opens the page, they see what the AI read.
WHAT THE AI DOES HERE
Five concrete things inside the workspace
None of them decides for a person. They prepare, they order and they leave a trail: publishing, classifying and granting permission remain human decisions.
workspace.draft()
From a conversation to a procedure in draft
It takes the thread where something got resolved — a chat, a meeting, an email — and turns it into a structured page: goal, steps, owners, exceptions. It stays as a draft, with an owner assigned.
› input
A 41-message thread about dispatching an order with stock reserved in two warehouses
› claude →
page: “Dispatch with stock in two warehouses” · draft v1
structure: goal · 7 steps · 2 exceptions · owner per step
proposed owner: operations lead
state: draft — not visible to the AI until it is published
LIVE SYSTEM · Plano Cero
This is what it looks like inside an AI-operated system — and how it is governed.
Plano Cero is an engineering and project firm with three offices. Its workspace holds procedures, minutes and databases, and its AI answers only from what is published and with the permission of whoever asks.
System modules
- 01Home and inbox
- 02Pages and procedures
- 03Databases
- 04Meeting minutes
- 05Sharing and classification
- 06Ask the AI
- 07Templates and history
Boosty Standard for Operating with AISimulated AI · demo data
The same data. Three ways of looking at them.
The “Procedures” database at Grupo Latitud is not duplicated per view: the table, the kanban by status and the review calendar read the same rows. The shape changes, not the source. And of those rows, the AI only reads the ones with a published version.
| Page | Owner | Space | Classification | Version | Status | Review |
|---|---|---|---|---|---|---|
| Andrés Beltrán | Operations | Internal | v4 | Published | 6 Oct | |
| Valeria Quintero | Sales | Confidential | v2 | Draft | 9 Oct | |
| Rubén Castellanos | Operations | Internal | v3 | Published | 14 Oct | |
| Marisol Prieto | Management | Confidential | v6 | In review | 16 Oct | |
| Valeria Quintero | Sales | Confidential | v7 | In review | 20 Oct | |
| Andrés Beltrán | Operations | Internal | v1 | Draft | 23 Oct | |
| Marisol Prieto | Management | Restricted | v2 | Published | 27 Oct | |
| Daniela Rivas | Sales | Public | v5 | Published | 30 Oct |
Warranty policy
Click a row, a card or a day to see its properties.
- Owner
- Andrés Beltrán
- Space
- Operations
- Version
- v4
- Last edit
- Sep 12 · 09:14
- Next review
- 6 Oct 2026
- Classification
- Internal
- Status
- Published
published, with a classification allowed in its space
The live version is one; the rest is history you can consult.
IN THE SYSTEM · DATABASES
The same rows, five ways to see them.
Table, board, calendar, timeline and gallery on the same database, with filters by office and status. Switching views does not copy the data: it rearranges it.
Real screenshot of the system · demonstration data
Who sees it, and how much of it the AI may read.
Permission comes from the space and the role. Classification decides whether that also enters the index the AI queries. They are two different things and the system does not mix them: a person may have access to a page the AI cannot read.
Access matrix
Three spaces, three roles. What is not here is not visible.
| Space | Management | Team lead | Staff |
|---|---|---|---|
| Sales | View and edit | View and edit | Read only |
| Operations | View and edit | View and edit | Read only |
| Management | View and edit | No access | No access |
Who sees this page
Wholesale price list · DR · Sales
Management · Team lead · Staff
These rules are not the editor’s invention: they come from the information-use policy — what gets classified, who approves it and how long it is retained. The workspace only enforces them.
Classification per page
Change the classification and watch what happens to the switch.
Internal: readable in the spaces where the person asking has permission.
Confidential: readable by the AI only inside its space; outside it, not even cited.
Restricted: never indexed. The switch stays locked and there is no exception by role.
Public: readable with no space restriction; it is material already shared outside.
Audit trail
Change a classification or a switch: every action leaves its line, with a name and a time.
IN THE SYSTEM · SHARE
The classification decides whether the AI can read the page.
Access by space and four classification levels: public, internal, confidential and restricted. «Visible to the AI» is not set by hand; it follows the classification, and every change stays in the history.
Real screenshot of the system · demonstration data
FROM CHAT TO PROCEDURE
What got resolved in a thread ends up being what the agent cites.
Five states of the same Grupo Latitud procedure: the conversation where it was decided, the draft the AI writes, a person’s review, the publication — pressed by a person — and the run where an agent uses it citing version and step.
WhatsApp · Operations and Commercial · 41 messages
Iván · Commercial · 09:12
Order 8841: 60 units reserved in the DR warehouse and 40 in VE. Does it ship complete or partial?
Marcela · Operations · 09:14
If it ships partial, the client pays two freights. Commercial authorizes that, not us.
Iván · Commercial · 09:15
Then we consolidate. How long does moving the 40 from VE to DR take?
Rubén · DR warehouse · 09:17
Three business days with the inter-entity transit. And the transfer note has to be issued before anything moves.
Marcela · Operations · 09:21
Rule: if the shortfall is under 30 % of the order, we consolidate. If it is above, it ships partial and commercial notifies the client.
Iván · Commercial · 09:23
Agreed. And who releases the reservation in the other warehouse?
Rubén · DR warehouse · 09:24
Operations, when the transfer is confirmed. If it is not released, the stock sits dead for a week.
Marcela · Operations · 09:26
This has happened three times already. We should just write it down.
The rule was decided here. It is written nowhere and nobody will open this thread again.
IN THE SYSTEM · FROM THE MEETING TO THE TASKS
The minutes already bring the decisions and the tasks with an owner.
The AI proposes each decision and task citing the note it comes from. Where it is not clear who does it, it does not assign it: it asks for a person. Confirming is up to the committee secretary.
Real screenshot of the system · demonstration data
SAME WORKSPACE, DIFFERENT KNOWLEDGE
What is written changes. How it is governed does not.
The same engine of pages, permissions and citations holds very different manuals. What stays is the rule: published, classified and with an owner.
Signals specific to the industry
HOW IT IS BUILT
Seven layers and a cut before the model.
Permission and classification are not an interface checkbox: they are evaluated in the database and in the index, before the model receives a single fragment. And a published version is never edited: it is frozen.
Block editor
What it does
You write in blocks — paragraph, table, checklist, embedded database — and every block has its own identity.
What it guarantees
A citation can point to the exact block, not the whole page. Whoever opens the link sees what the AI read.
Structured JSON
What it does
The document is stored as a tree of blocks with their type and attributes, not as a file or loose HTML.
What it guarantees
It can be read in parts, compared across versions and re-rendered on any surface without losing structure.
Postgres with per-space RLS
What it does
Every row carries its space, its classification and its owner. The access policy is evaluated inside the database, row by row.
What it guarantees
A query without permission returns no rows. There is no application-side filter for anyone to bypass.
Search index
What it does
Full text plus embeddings over the same content, indexing only pages with a published version.
What it guarantees
Drafts never enter the index. Neither does restricted content: it is not hidden in the answer, it was never indexed.
The filter happens here
The model does not query the database: it receives the fragments that survived the space permission, the classification and the published state. What did not pass this cut does not exist for the AI.
Claude with citations
What it does
It receives only the fragments that passed the cut and writes the answer over them, with page, version and block.
What it guarantees
No fragment, no answer: it says the page is not published and names its owner.
Snapshot versions
What it does
Publishing freezes the whole tree into a new version. Later edits work on a separate draft.
What it guarantees
The version cited yesterday opens the same today. History is never edited or deleted; it is only appended to.
SSO and roles
What it does
Identity comes from your corporate directory; the role defines which spaces each person enters and at what level.
What it guarantees
Whoever leaves the company loses access in the same place where their account is disabled, with no separate list to maintain.
All of this lives in your database. Content is used as context for the query, not to train a model.
IN THE SYSTEM · ASK THE AI
It answers from what is published. What is not there, it says.
Every answer cites the page, the version and the section. If only a draft exists, it says it is not published; if the page is restricted for whoever asks, it says they have no access, not that it does not exist.
Real screenshot of the system · demonstration data
WHAT WE MEASURE
Adoption and knowledge health, not projected return.
Four indicators that tell whether the workspace is alive: how much is published, how many questions are answered with a source, how long finding something takes and who actually uses it. None of them promises a result: they describe a state you review.
Published vs draft pages
128
of 169 pages in the workspace
What sits in draft is not read by the AI. This number says how much knowledge still lacks an owner to publish it.
Questions answered with a source
612
of 708 questions in the period
The 96 without a source are not a failure: they are the work list. Each one names something still to write or to publish.
Time to find (median)
27 s
from the question to opening the block
We track the median, not the average: one rare slow search should not hide how the rest of the team is doing.
Use by role
88
of 112 people with access, this week
88 / 112 people
A workspace used by one area only is not a single source of truth: it is someone’s folder. By role you see where publishing is missing.
These are the indicators we watch, not a promise of a result. The numbers above are the state of a Grupo Latitud demonstration workspace: in your operation the starting point is set by the assessment, not by this board.
IN THE SYSTEM · A DAY IN THE WORKSPACE
Home, inbox, procedures and databases with each person’s permissions.
What awaits your review, mentions, the published procedure and the project portfolio. Restricted content disappears from the tree, the search and what the AI reads for anyone without access.
Real screenshot of the system · demonstration data
WITH WHAT YOU ALREADY USE
The workspace does not live alone
It connects with the sources where the work actually happens. What comes in gets classified; what goes out gets cited.
Claude · Anthropic
EngineDrafts, answers with citations and proposes the classification. It never publishes on its own.
Google Workspace
Email, calendar and documents as a source: a meeting ends as a page with decisions and tasks.
WhatsApp Business
The thread where something got resolved becomes a draft procedure, with its owner.
Supabase · Postgres
Pages and their versions live in your database, with per-space permissions at row level.
Make · n8n
Publishing a page can trigger notices, tasks or the update of an agent.
Kommo CRM
The published sales playbook is the one the agent uses when preparing a follow-up.
Frequently asked questions about collaboration systems
It depends what you use them for. If they work for writing together, keep them. This gets built when knowledge has to live alongside fine-grained permissions per space and role, information classification, a full audit trail and an AI that may only read what is published and allowed. That is where the generic tool falls short, and that gap is the reason to build it for you.
It reads pages with a published version, inside the spaces where the person asking has permission, and only the classifications the policy allows. Drafts are not indexed. Restricted pages are not indexed. And every answer arrives with the page and version it came from: no source, no answer.
Every page has one live version and a full history behind it. Publishing creates a new version; the history is not deleted. Retention is defined by data type in the information policy, and the system enforces it: what is kept, for how long, and who can recover it.
Yes, and it is part of the work. What exists is imported, given an owner and a classification, and then you decide what gets published and what gets archived. Migration is not copying folders: it is deciding what the single source of truth is for each thing. That comes out of the assessment.
No. Content is used as context for the query, not for training. Documents live in your database, with row-level permissions, and the model only receives the fragment needed to answer the question it was asked.
It is defined in the assessment. The model has three parts: setup, a monthly fee and a variable cost per processed volume. The scope — which spaces come first and what gets migrated — comes from what we see in your operation, not from a catalog. Book 30 minutes and we give you the range in writing.
A WORD FROM THE FOUNDER
“The sentence I hear most is not “we need AI”. It is “that is in a chat somewhere”. Company knowledge lives in three people’s heads and in conversations nobody opens again.”
There is no artificial intelligence without information. If your operation has no place where the live procedure is written down, with an owner and a date, no model fixes that: it will just answer what it does not know with more confidence.
That is why this workspace does not start with the editor. It starts with three questions: what is the single source of truth, who may see it, and what happens when it changes. The editor comes after, and the AI later still — once there is something published worth reading.
If you want to see where what your team needs is written down today, book 30 minutes with me. We take a real procedure of yours and put it in order together.
Gabriel Montiel
CEO · Boosty Digital
START
Where does your company knowledge live today?
Book a 30-minute assessment. We look at where what your team needs is written down, what is still unpublished, and what the AI could read with permission.