Product · AOS.work
One place where people, agents and apps work from the same context, the same standards and the same guardrails. Start a task, continue a conversation, use an authorised capability, receive the output. Every run traceable, sandboxed and cancellable.
One entry
New Task, Search, Explore, Memory, Scheduled Tasks and Task Outputs in one left navigation.
One runtime
Built on Basil Harness. Every run is a traceable, sandboxed, cancellable unit of work.
One gateway
Protected by Shieldon. Policy, security and audit for every model call.
Chapter 01 · Overview
AOS.work is the place where a company's context, its ways of working and its files and data come together, so that an agent can do real work under governance. Three modules, one interface.
Chapter 02 · Context & Memory
Memory holds what the agent should know about a person for the long term. The User Profile shows all of it, every edit creates a new revision, and nothing is inferred to be stale.
- #1
Prefers a short weekly summary with a separate risks section.
- #2
Works in the Q3 launch Workspace most days.
- #3Edit
Delivers reports as a .pptx deck in Library / Sales.
- #4
Uses CNY for every budget figure.
Update History
- r12Consolidation · nightlyyesterday, 02:00
- r11Manual edit · currencylast week
Revisions, not overwrites
Editing Memory creates a new Memory Revision. Update History lists them all and any one can be restored. A restore is itself a new revision, so the history is never lost.
Consolidation on request or on schedule
Ask in chat, or let a nightly Scheduled Task merge duplicates, apply corrections and forget markers, and drop expired notes. It never decides on its own that a fact is out of date.
Context tags in the composer
A Workspace, attached files, Library files and Base views become context tags on the task. Works and Skills are tagged the same way. What the agent sees is exactly what you picked.
Outputs that trace back
Task Outputs collect the final files and URLs of every task, with Back to Original Message on each one. Resource links open only after a live permission check.
What travels with a task
The agent sees exactly what you picked.
Compare this quarter's pipeline with the plan and flag anything more than ten percent behind.
Back to Original Message
Anything missing is asked for explicitly. Nothing is guessed.
Chapter 03 · Skills & Works
A Skill is a reusable capability for the agent. A Work is a package of Skills for a role or a whole job. A Connector reaches an external system. Explore is where people find them, admins release them, and versions stay immutable.
- 01
Discover in Explore
Skills, Works, Connectors and Apps with their current status. Describe what you need in chat and the built-in Explore Skill points you to the right published resource.
- 02
Bind the Connector yourself
A Work declares its Connector Requirements. You pick the Connection explicitly; nothing is chosen or switched for you. One Connection per person per Connector.
- 03
Test before the first run
A Connector can be called only after Test Connection passes. An invalid binding puts the run into Awaiting Authorization instead of failing quietly.
- 04
Released as an immutable version
Resource Hub validates the draft and publishes a version that never changes. Explore Management releases it to roles, pinned to a specific version or always the latest.
Finance Close Kit
monthly-report
Document toolkit
ERP Connector
contract-review
Sales Prospecting Kit
Skill
One reusable capability. A Skill card can hold a collection of skills, each with its own toggle. Enabled user-level Skills load by default.
Work
A package of Skills for a role or a complete job. Select it in the composer and its Skills load together.
Connector
A capability of an external system. No authorisation, a static credential through a secure form, or OAuth 2.0.
Every resource carries its status in Explore
- Enabled
- Unauthorized
- Unconfigured
- Needs Reauthorization
- Connection failed
- Update Available
- Policy Locked
Chapter 04 · Workspace & Apps
A Workspace scopes the files and tasks of a piece of work. Library and Base are apps you add and authorise in Explore, after which they appear as top-level entries. Scheduled Tasks run on time, and Inbox reports back.
Workspace
Explicit or implicit, a Workspace scopes files and tasks so continuous work stays continuous.
Library
My Space, Team Space, Recent, Shared with Me, Favorites, folders, sharing and trash. The file system and the source of truth.
Base
Tables, fields, records and views. Filter and group, or ask a task to return a query view.
Scheduled Tasks
A fixed message on a one-time or recurring plan, with Workspace, optional Work, Skill or model, time zone and date range.
Inbox
Explore updates, scheduled-task results and administrator messages in one stream, in the top bar.
Apps
Added and authorised in Explore. An app can declare page entries, Open Origin System and admin-managed fields.
Every trigger creates an independent Task and an immutable Run Record
- 01Pending
- 02Running
- 03Succeeded
- 04Failed
- 05Skipped
- 06Awaiting Authorization
How the modules fit together
Context tells the agent what is true. Skills tell it how work is done here. The Workspace gives it the files and data to work on. Basil Harness executes the run in a sandbox, and Shieldon governs every model call along the way.
Context & Memory
Workspace, files, views and Memory become the task's context.
Skills & Works
The right Work loads its Skills and bound Connectors.
Workspace & Apps
Library files and Base records are read and written in scope.
One governed run
Traceable, sandboxed, cancellable
- run.queued
- run.started
- tool.started
- tool.completed
- context.checkpoint.created
- run.completed
Basil Harness
Every run is a traceable, sandboxed, cancellable unit of work.
Shieldon
Policy, security and audit for every model call.
Book a demo and we will set up a workspace with your knowledge, your standards and your guardrails.
Built on Basil Harness · Protected by Shieldon