Agentic AI Website Development and Content Management Platform
A multi-tenant platform where each client gets a login-gated AI assistant that builds and manages their website and content by chat — with role-based permissions and human approval on every change.
One console
provision and configure a new client assistant in minutes
Read→Create
granular content permission ladder enforced on every write
100% audited
every automated change recorded in an immutable log
Per-tenant
isolated assistants, encrypted keys and metered billing
The brief
Digital Services Agency
Web & Content Management (SaaS)A multi-tenant platform where each client gets a login-gated AI assistant that builds and manages their website and content by chat — with role-based permissions and human approval on every change.
The challenge
What needed to change
Editing a website still means raising a ticket, waiting for a developer, or wrestling with a content system that non-technical staff find intimidating. Our client wanted to give every one of their own customers something better: a conversational AI assistant that could actually make the change — write the copy, update the page, even ship the code — rather than just advising on how to do it. Doing that safely across many separate customers was the hard part. Each customer needed their own isolated assistant, their own connected systems and keys, fine-grained control over who could edit versus publish, a complete audit trail of every automated change, and per-customer usage metering and billing — all provisioned and managed centrally, without standing up new infrastructure for each one.
Key requirements
What we needed to deliver
- Give each client a private, login-gated agentic AI assistant that can both manage website content and make code changes — not just answer questions.
- Isolate every client as its own tenant, with separate users, connected systems, encrypted keys and data.
- Enforce fine-grained, role-based permissions on every action, including a content ladder from read through edit, publish, delete and create.
- Require human approval and keep a full audit trail for every change the assistant makes.
- Compose each assistant from modular skills — content management, code, web research, stock and generated imagery — switchable on per client.
- Stay content-system agnostic, connecting to any headless CMS through an adapter rather than being tied to one.
- Meter token and third-party usage by user and session, and bill it back with a configurable markup, by invoice or card.
- Let the operator provision and configure a new client assistant — model, prompt, keys, connections and billing — from one master console in minutes.
The solution
How we solved it
iCentric turned a bespoke website assistant it had built for a single site into a multi-tenant SaaS platform. The principle is simple — one client, one assistant, one tenant — but the assistant is genuinely agentic: it plans and carries out real work against the client's live systems through a tool-use loop, rather than returning instructions for a person to follow.
One assistant per client, composed from skills
Each client's assistant is assembled from modular skills — managing content, editing code, live web research, sourcing stock imagery and generating images — each bundling its own tools, prompt guidance and credentials. Content and code are enabled by default; the rest switch on the moment an admin adds a key. It connects to the client's content management system through an adapter, so the platform isn't locked to any single CMS. Because skills are registry-driven, new capabilities roll out to every assistant at once without rebuilding any of them.
Safe by construction — permissions, approvals and audit
Nothing the assistant does goes unchecked. A capability model separates the operator's platform administrators from each client's own team admins, and a granular content permission ladder — read, edit, publish, delete, create, plus a separate permission for code — is enforced on every write. Sensitive actions are gated behind explicit human approval, and every change the assistant makes is recorded in an immutable audit log, so there is always a complete, reviewable trail of what was done, by whom, and when.
A multi-tenant engine with central control
Each assistant turn is handled as a queued job: the request is authenticated, a run is created, and a long-lived worker composes that tenant's assistant from its enabled skills and decrypted credentials, runs the AI loop, and streams progress back to the browser in real time. Per-tenant keys and credentials are held under strong encryption. From a single master console the operator can stand up and configure a new client assistant — model, prompt, connected systems and billing — in minutes, and watch usage, cost and system health across every tenant from one dashboard.
Usage metering and billing built in
Every token and third-party call is metered per client, per user and per session into a usage ledger, and each client's price is calculated from real cost with a configurable markup — billed by enterprise invoice or card. Clients get their own self-service area to manage their team and roles, review their members' activity, and see exactly what they are spending.
The outcome
The impact
The platform turns website and content management into a conversation. A client's non-technical team can ask for a change in plain language and watch the assistant carry it out — updating content or shipping code — with the right people approving anything sensitive and every action logged. For the operator it converts a one-off, single-client assistant into a repeatable product: a new client can be onboarded with their own fully configured, isolated assistant from one console in minutes, with usage and billing handled automatically. It's a blueprint for delivering agentic AI as a governed, multi-tenant service rather than a bespoke build each time.
Get in touch today
Book a call at a time to suit you, or fill out our enquiry form or get in touch using the contact details below