FPT AI Agents EOS - A platform where people and AI Agents do real work together, under one shared set of rules and controls.
How the business changes
Most companies today work like this: the work lives in inboxes, in meetings and in people’s heads. Know-how belongs to individuals, so when someone leaves, the knowledge leaves with them. Managers only see results once the quarter has closed. Improvement happens project by project, review cycle by review cycle. EOS changes those four things:“Operating System” here does not mean a computer OS. It means the shared coordination layer of the business: one place that holds the knowledge, the processes, the permissions and the logs of every Agent.
How this differs from a chatbot and from an ordinary AI Agent platform
All three are usually lumped together as “AI”, but they are fundamentally different.Five core capabilities
Governance and Knowledge are the foundation - every Agent runs beneath them. Process, Productivity and Analytics are the value layer. The order matters: without policy there is nothing to supervise, and without reliable knowledge the Agent gives wrong answers in every process downstream.1. Governance
Five steps, from the basics to the most advanced: define policy, grant permissions by context, monitor and detect violations, enforce and route to approvers, audit. What sets EOS apart is that policy is not an internal document; it is wired straight into the execution path. A customer-care Agent may draft an email but not send it. An onboarding Agent cannot read payroll data. Violations are blocked at the moment they occur, not discovered after the message has gone out.2. Knowledge
This goes beyond document lookup: comprehension, retrieval, citation, analysis, multi-source reasoning, and finally the creation of new knowledge. A question such as “Can I sign a freelancer contract, and what is the withholding tax rate?” has its answer split across two documents owned by two departments. The Agent brings them together in one conversation and cites each part from its own source.3. Process - Automation and process redesign
Two operating modes. Follow the procedure: you define each step, the Agent orchestrates it precisely and flags exceptions. Reach the objective: you define only the objective and the principles, and the Agent finds the fastest route within the boundaries you allow. After a period of operation the system points out the steps that are no longer needed - a seven-step onboarding process may come down to three.4. Productivity
Everyone’s work splits into two kinds: work that needs judgement, and work that only needs execution. EOS does not replace judgement; it makes judgement better by preparing the data, drafting the first version and pointing out the risks. It takes over the second kind entirely. For example, a QA manager who could previously review only 2-3% of 10,000 calls a week now has 100% of conversations scored automatically, and receives a list prioritised by actual risk.5. Analytics
Four levels: descriptive, diagnostic, predictive, prescriptive. Most organisations are stuck at the first, not for lack of data but for lack of time to turn data into insight. For example, loan approval rates fall from 78% to 62%. Instead of three to five days of manual analysis, the system shows within minutes that the drop is concentrated among applicants under 30, on short-term consumer loans, handled by two new staff members.Notable features for the enterprise
Build Agents in natural language, no code required
Describe what you want the Agent to do and the system drafts it for you to refine. Attach skills, tools, knowledge and guardrails right in the Agent builder. Complex work can be split across several supporting Agents that work together.One shared knowledge source, answers with citations
Company documents go into one place, and every permitted Agent draws on the same source. Answers come with specific citations, so users know which document the information came from. Update a policy once and every related Agent reflects the change immediately.Guardrails and human approval
You set the boundaries for a single Agent or for the whole workspace: topics that are off limits, personal data that must be masked, risky actions that need human approval before they run. Change one shared policy and a whole set of Agents changes with it.Connects directly to the systems you already use
Agents read and write directly in your CRM, ERP, HRIS, SharePoint or Jira through connectors and MCP, so data is never copied out of its system of record. Agents serve customers on the web, Zalo, Teams, email and the voice contact centre.Test first, publish after
Every change stays in a draft until you publish. There is a test tab for trial runs and versioning to roll back when you need to, so the Agent currently serving customers is never affected by the version you are editing.Logs and compliance reporting
Every action an Agent takes is recorded immutably. You can replay exactly what happened in any session, and export a compliance report on request.Organisation and cost governance
Permissions by role, by workspace and by individual Agent. Data isolation between business units. Cost, quality and performance for every Agent in one place.The knowledge loop: the more you use it, the better it gets
This is what separates EOS from a collection of disconnected tools. Daily work produces new knowledge, and that knowledge flows back into the platform:- Work - people and Agents run thousands of tasks a day.
- Capture - exceptions, better ways of working and customer insight are kept rather than lost.
- Confirm - AI proposes, a domain expert confirms. An AI suggestion is a suggestion, not a conclusion.
- Enrich - record the reasoning, the exceptions and the places where teams disagree.
- Publish - once approved, the knowledge enters the shared library.
- Reuse - every person and Agent with permission can draw on it.
Who EOS is for
EOS is designed for the work of every department: Finance, Sales, HR, Operations, Customer Service, Risk and Compliance.
Where to start
The documentation is split in two, matching the two ways people use the platform.FPT AI Workspace
For end users: chat with Agents, work with Artifacts, get Agents from the Agent Marketplace and set up personal connections.
FPT AI Console
For builders and administrators: create and edit Agents, Instructions, Model, Skills, Connectors, Guardrails, publishing and organisation governance.
1
Build your first Agent
Go to the Console, describe in natural language what you want the Agent to do, then refine the Instructions and choose a Model.
2
Load knowledge and connect your systems
Attach your business documents, add Skills and Connectors to your CRM, ERP or SharePoint so the Agent can do real work.
3
Set the boundaries
Configure Guardrails - what the Agent must not do, and which actions need human approval.
4
Test, then publish
Try it with the questions people actually ask, then publish and choose your deployment channels.