How FactoryOS Decides Who Sees What

How FactoryOS Decides Who Sees What

Most systems split access control into two separate mechanisms: one decides which pages a person can open, the other decides which data they can see. FactoryOS uses a single engine for both. The same three-tier cascade that decides whether someone can open the Workflow editor also decides whether their assistant can see the HR channel — and the same screen configures the two side by side.

One Engine, Two Surfaces

The permission engine governs two things at once. Features are the pages, tools, and capabilities a person can use — the Workflow editor, the personal assistant, Persona Studio, the prompts library, the dashboard widgets, the ability to manage other users. Channels are the partitions of data the brain can see for that person — the shared company channel, project-scoped channels, the user's own private channel.

The permissions screen places these as two tabs — Permissions and Channels — and the same cascade resolves both. That matters because feature access and data access are usually entangled in messy ways. Splitting them into separate tabs but resolving them with the same logic keeps both explicit.

How a Decision Gets Made

A permission decision walks three layers, top to bottom:

1. User override. A setting placed on this individual person, if one's been set. 2. Their role. The role they belong to — default or custom — if it has an opinion on this permission. 3. System default. The baseline that ships with FactoryOS.

At each layer the answer is one of three states: allow, deny, or inherit. The first layer with an allow or deny wins, and the cascade stops there. Layers that say inherit defer to the next layer down.

A user with no personal overrides falls through to their role; a role with no opinion on a specific permission falls through to the system default. That keeps individual customization light — most people get the role they're given, and the role does the work — while still allowing exceptions when one is warranted.

Seven Bedrock Roles

FactoryOS ships with seven roles that cover most organizations out of the box:

- Owner — Immutable, the top of the chain, typically the CEO/Founder. - System Admin — the IT-side role, manages users and system settings. - Executive — the C-level leadership tier. - Manager — the middle-management tier. - Member — the typical operational employee. - HR Officer — handles the people side of the business. - Auditor — read-only oversight.

Custom roles can be added on top — "sales floor," "contractor," "intern" — each starting from a permissions baseline you set, and individual users can then override specific permissions on top of whatever role they belong to.

Why Channels Never Bleed

The brain's data is partitioned into channels. The company-wide channel holds shared knowledge — policies, organization-wide announcements, the structure of the company. Project channels hold work scoped to specific clients, deals, or initiatives. Each person also has a private personal channel that nobody else can see.

Channel (data) access flows through the same cascade as features. Default roles get default channel access; custom roles get whatever channel set you defined for them; individual users can be granted or denied specific channels on top of that. The HR channel doesn't bleed into operations chat because operations roles don't have HR-channel access — and the cascade makes that not a checkbox someone could forget to tick, but the baseline behavior.

The personal channel is the strictest case. Nothing at the role layer can reach it; nothing at the system-default layer can reach it. Only the user themselves sees it, and the [assistant attached to them](how-your-personal-assistant-stays-yours).

Every Page Can Be Gated

The granularity is per-feature. The permissions UI groups capabilities by area — Administration, AI, Dashboard, Tools, Assistant — and inside each area, individual pages and capabilities are independently controllable. The Workflow editor, the Manifold tool, Persona Studio access, the personal-assistant chat in the header, the prompts library, the activity heatmap on the dashboard, the ability to manage other users' permissions — each is its own toggleable line in the matrix, set independently per role and per user.

The same matrix flips the view between By Role and By User. By Role shows what a role can do across every feature; By User shows what a specific person can do after their personal overrides resolve on top of their role. Both views read off the same underlying cascade. All from one simple interface.

That's the brag. There isn't a feature in the system that can't be turned off for a role or a person, there isn't a channel that can't be walled off the same way, and the answer to "can this user do this?" is always one of the three states with a single source you can trace back through the layers.

Recent Articles

How FactoryOS Builds Charts and Diagrams

Ask FactoryOS for a chart, diagram, table, or image and it renders on your own hardware. Eight local renderers on one canvas, nothing sent to a cloud service.

How the Knowledge Graph Remembers Over Time

FactoryOS's brain remembers what changes over time. Confidence, resilience, and expiration dates make its graph behave more like real memory.

Visual Workflows That Run on Their Own

Draw a workflow on a visual canvas, send it to FactoryOS's runtime, and watch it fire on its own. AI calls, schedules, button triggers, and a full run log.

How Your Personal Assistant Stays Yours

FactoryOS's personal assistant talks only to you. Daily briefing, contact lookup, a character you customize, and a voice you pick, all on your hardware.

Where Your Assistant Gets Its Face

Persona Studio is where you sculpt your FactoryOS assistant's look. Chat to refine, lock a seed, pin favorites, and let the system restage the rest.

How FactoryOS Listens and Speaks

Voice is baked into FactoryOS, not bolted on. Local speech engines, swappable models, downloadable voices, custom voice training, and admin-set menus.

How FactoryOS Pilots a Real Browser

FactoryOS hands the model a real Chrome browser, strips the page noise, and carries out a research mission you describe in plain language.

How FactoryOS Retrieves the Right Context

A model is only as good as the context you give it. How FactoryOS stacks keyword, vector, fusion, reranking, and embeddings to retrieve the right passages.

Popular Articles

What Separates an AI OS from a Wrapper

There are three things people call AI tools. Here is what separates a chat wrapper, an agentic tool, and a full AI operating system

Inside the Factory Knowledge Graph

Some answers live in the relationships between documents, not any one of them. How the Factory Knowledge Graph stores time-aware facts you can reason over.

How FactoryOS Retrieves the Right Context

A model is only as good as the context you give it. How FactoryOS stacks keyword, vector, fusion, reranking, and embeddings to retrieve the right passages.

How FactoryOS Pilots a Real Browser

FactoryOS hands the model a real Chrome browser, strips the page noise, and carries out a research mission you describe in plain language.

Why Heavy Load Means Delays Not Crashes

A single machine has a finite GPU. How a queue runner and priority scheduling make heavy load show up as a delay, not a crash or a surprise bill.

How FactoryOS Builds Charts and Diagrams

Ask FactoryOS for a chart, diagram, table, or image and it renders on your own hardware. Eight local renderers on one canvas, nothing sent to a cloud service.

How the Knowledge Graph Remembers Over Time

FactoryOS's brain remembers what changes over time. Confidence, resilience, and expiration dates make its graph behave more like real memory.

Visual Workflows That Run on Their Own

Draw a workflow on a visual canvas, send it to FactoryOS's runtime, and watch it fire on its own. AI calls, schedules, button triggers, and a full run log.

Other Categories