Agent Roles with Skills
Create role presets such as researcher, assistant, sales assistant, finance helper, content producer, workflow builder, and auditor. Each role can carry its own skills, habits, tools, and working style.
99Pages Agentic OS lets one AI agent work across roles, skills, tools, workflows, devices, memory, and remote control.
It is not another chatbot and not an LLM model. It gives your agent a working environment: model routing, role skills, task memory, approval controls, local software access, scheduled automation, and device-aware execution.
Save up to 95% of tokens spent on startup and context reloading compared with naive full-context loading.
Users do not need to babysit a browser tab. They give the instruction, choose the permission level, and let the agent work through the right role, model, tool, and device.
Create role presets such as researcher, assistant, sales assistant, finance helper, content producer, workflow builder, and auditor. Each role can carry its own skills, habits, tools, and working style.

Bind a main model, choose fallback models, and allow task-based switching for chat, research, automation, image work, video work, coding, and review.
Turn repeatable work into structured flows with approvals, tool calls, checkpoints, retries, and execution receipts.

Plan commitments, schedule work, track open loops, and use structured tables for operational data, handoff lists, local records, and repeatable job inputs.
Agentic OS is designed around a local Keybox for sensitive service credentials, provider access, private account handles, and server access profiles.
The agent can use an approved credential handle to complete a task, but it cannot reveal the private value in chat. This lowers credential-leak risk while still allowing real automation.
Start jobs from the desktop app, a web account, or Telegram. The agent can keep running on a server, pick the right local device when it comes online, and report back without requiring you to stare at the screen.
For local software, browser accounts, home-network browser work, rendering, Office files, and device-specific tasks, Agentic OS can route the job to the machine that actually has the right environment.



Premium users can activate a 90-Day Memory Vault for distilled working memory: role habits, tool preferences, task patterns, approval style, successful workflows, and reusable operating know-how.
The memory pack is designed for backup, migration, and team deployment. Move it to another device, restore it after reinstalling, or use it as a starting point for a team agent without backing up full raw chats by default.
Export what matters: preferences, skills, routes, task playbooks, tool habits, and decision patterns.
That means a trained agent can feel familiar on a new machine, or become a cloneable starting point for a team.
Start free, choose a 30-day plan, then add Time Pack hours when you want more automation capacity. One control channel can coordinate multiple registered devices.
$0 / 30 days
For trying the desktop command layer.
$9.90 / 30 days
For personal automation and remote control.
$29.90 / 30 days
For heavier workflows across more devices.
$79 / 30 days
For memory migration and cloneable working style.
Time Pack hours are optional top-ups for extra automation capacity. Extra registered devices can be added as the workspace grows.
Rates vary by country, state, age, award, sector, and law. These examples are a simple business reference for comparing a software subscription with manual labor cost.
| Market | Public minimum wage reference | 99Pages Premium |
|---|---|---|
| Malaysia | RM1,700 per month | $79 / 30 days |
| United States | $7.25 per hour federal rate | About 11 federal minimum-wage hours |
| United Kingdom | £12.71 per hour National Living Wage | Software subscription, not payroll |
| Australia | A$24.95 per hour national minimum wage | Software subscription, not payroll |
The user benefit is simple: faster starts, lower token waste, stronger continuity across sessions, and an agent that feels trained by your way of working.
Potential reduction in startup and reload token spend compared with naive full-context loading.
99Pages Agentic OS is based on the 99Pages concept of Governance Unified Field Theory and Ultralight Governance.
The project is supported by the registered works Enterprise Operating System and Enterprise Intelligent Role Reconstruction Guide, registered with the National Library of Malaysia.
The developer owns 100% of the development copyright, publishing right, distribution right, trademark use right, operating right, and final interpretation right.