Platform architecture

One platform inside your company, three layers

The GoEdge platform is installed inside your company. The bottom layer controls who can use what and what has been done; the middle layer lets AI work within rules; the top layer is the applications that do the work for you.

Every connection is started from the inside, by the internal server and the computers that run tasks, so nobody outside can connect in to the platform. See Security and deployment (Chinese).

Management and protection layer

Whether AI can be trusted in a business depends on this layer.

Identity and permissions

Sign in with company accounts, connected to your existing account system. Five roles each have their own job: administrator, skill maintainer, task designer, operator and auditor. AI always acts on behalf of one person and never has more rights than that person.

Secrets vault

Passwords exist only inside the platform. Tasks and logs show a placeholder only; the real password is used at the moment it is needed and never appears on screen, in logs or in an AI conversation.

Audit log

Deployments, approvals, permission and setting changes are written as a chain of records, each linked to the one before. If someone changes or deletes one afterwards, it can be detected.

AI brain and assistants layer

Make AI affordable, controllable and able to collaborate.

Model management

Your own AI models and external AI services are allocated from here, with rules by purpose, software and department. External services come with data rules and a fallback; people and departments can have daily, weekly or monthly usage limits.

Skill library

Turn know-how about procedures into “skills”. A new skill is proposed, reviewed, then published, with version history and comparison; a proposer cannot approve their own proposal; skills meant for one department appear only there; and a mistake can be rolled back.

Tasks and approvals

Work is queued with priorities, schedules and notifications (Teams and Slack can be connected). Anything affecting production always needs approval, and the approver cannot be the requester. Timed-out work is rescheduled automatically and failures can be retried.

Developer tool access

Developers can use their favourite AI development tools to build and publish internal tools through a single entry point the platform provides. Permissions follow the actual user, and the tools never get the platform's passwords.

Let applications be used by AI assistants too

Applications on the platform can open their capabilities to the platform's AI assistants. For example, ManuFlow opens up scheduling, so an AI assistant answering a question or doing a task can use the schedule and delivery dates. The AI assistant still acts as the real user and never has more rights than that person.

Who does what: five roles

The platform writes “who can do what” into permissions instead of relying on verbal agreement.

RoleTypical personWhat they can do
AdministratorCustomer ITManage accounts, the computers that run tasks, notifications and usage limits
Skill maintainerIT or a consultantReview and publish skills, roll them back, decide which departments can see them
Task designerProcess ownerDesign task templates, decide whether approval is required and when to run, and approve other people's tasks
OperatorSales assistants and similar rolesPick a template and submit a task, watch the live screen, and give the AI instructions or flag mistakes
AuditorSecurity or internal auditView and verify audit logs, and replay the screen and report from a task

Want to see how the platform fits your environment?

Book a platform demo and we'll explain it against your systems and security requirements.