AI Engineering
AI that plugs into your existing platforms
We connect Claude, ChatGPT, Gemini and Copilot to your CMS, DAM, rules engine and internal APIs through secure MCP servers. Production-grade, observable, and governed for the EU AI Act.
Enterprise-grade from day one
What every AI integration we deliver comes with.
Claude, ChatGPT, Copilot, Cursor, Gemini, Mistral and more connect to the same MCP server. Switch models without rebuilding.
Each user authenticates with your SSO, and every tool only sees what that user is allowed to see.
Who asked, which tool ran, what changed. Logged and ready for your EU AI Act register.
On your real systems, connected to the assistants your team already uses.
What we build
AI work anchored in the enterprise platforms we have delivered for years.
We expose your CMS, DAM, rules engine or internal APIs as typed tools that any MCP client can call. Built on Cloudflare Workers with auth, rate limits, observability and interactive MCP App views.
llms.txt, Markdown versions of every page, RFC 8288 discovery headers, a public MCP endpoint and an AI crawler policy that does not silently block the agents you want.
Agents that model content types, populate sites, translate and publish through the Jahia GraphQL API, with a human approval step before anything goes live.
The model reads documents and extracts facts; IBM ODM makes the decision. You get language understanding with deterministic, auditable rules, which is what regulators expect.
Auto-tagging and metadata enrichment, usage tracking, and agents that search, collect and manage Bynder assets through MCP.
Claude Code and agent skills tailored to your stack (Jahia modules, Vaadin views, ODM rules), so your developers ship with your conventions baked in.
How it works
The Model Context Protocol (MCP) is the open standard that lets AI assistants call external tools. Instead of building a separate chatbot, we give the assistants your people already use a safe, well-defined way into your systems.
One MCP server works with Claude, ChatGPT, Gemini, Copilot and Cursor. You keep control of what each tool can read or change, and you can swap models without rebuilding the integration.
Agents ask, tools answer
The model never touches your database. It calls narrow, typed tools you control.
Runs at the edge
MCP servers on Cloudflare Workers: no servers to patch, global latency, per-IP rate limits.
Observable by default
Every tool call is logged with its purpose, so you can see what agents actually use.
Try it in 30 seconds
Our AI stack
Model-agnostic by design. We pick the model per task, not per vendor contract.
Models
Claude, OpenAI GPT, Google Gemini, and open-weight models for on-premise or EU-hosted needs.
Protocols
MCP (Streamable HTTP), MCP Apps for interactive results, OAuth 2.1 for authenticated tools.
Runtime
Cloudflare Workers, Agents SDK, Durable Objects, Workers AI and Vectorize for retrieval.
Enterprise systems
Jahia GraphQL and provisioning APIs, IBM ODM decision services, Bynder and ImageKit DAM.
Developer tooling
Claude Code, custom agent skills, Cursor and VS Code agent mode, CI checks on every change.
Governance
Provenance registers, human-in-the-loop review and EU AI Act Article 50 transparency.
See it working
Everything we sell, we run ourselves first.
How we made this site readable by AI agents: llms.txt, Markdown twins, discovery headers and a public MCP server.
A full Jahia demo site built by agents from specs to video, with the content created through GraphQL.
What changes for content teams when AI writes or edits published material, and how to stay compliant.
How we work together
From first call to a governed production system, in clear steps.
1. AI readiness audit
We map the systems, data and risks: which tools agents need, what they may change, and where human review is required.
2. Prototype MCP in two weeks
A working MCP server on your real data, connected to the assistants your team already uses.
3. Production hardening
Authentication, rate limits, logging, tests and deployment to Cloudflare or your own infrastructure.
4. Governance and handover
AI Act classification, provenance register, documentation and training so your team owns it.
Governance is built in, not bolted on
Every AI system we deliver ships with risk classification, provenance tracking and human-in-the-loop review where it matters. For regulated teams in banking, insurance and healthcare, our dedicated AI Governance Engineering service goes further with Article 50 compliance architecture and CI/CD compliance gates.
Explore AI Governance EngineeringFrequently Asked Questions
What teams usually ask before starting an AI project with us.
What is MCP and why not just build a chatbot?
MCP (Model Context Protocol) is an open standard for connecting AI assistants to tools and data. A chatbot is one more interface your people must adopt; an MCP server brings your systems into Claude, ChatGPT, Copilot or Cursor, which they already use, and it works across all of them.
Is our data sent to the model provider?
Only what a tool returns for a given request. You decide what each tool exposes, read-only or not. For sensitive workloads we can use enterprise model agreements with no training on your data, EU-hosted endpoints, or open-weight models on your infrastructure.
Which models do you work with?
Claude, OpenAI, Gemini and open-weight models. MCP keeps the integration model-agnostic, so you can change models later without rebuilding.
Can it run on-premise or in the EU only?
Yes. MCP servers can run on Cloudflare (with regional data controls), in your cloud, or on your own servers next to Jahia or IBM ODM.
Does it work with Jahia 8 and our existing modules?
Yes. Agents work through the Jahia GraphQL and provisioning APIs, so no core changes are needed. Publication can stay behind your existing workflow and approvals.
How does this fit with the EU AI Act?
Most agent integrations are limited-risk systems with transparency duties. We classify each use case, document it in a provenance register and add human review where required. See our AI Governance Engineering service for regulated contexts.
How long does a first project take?
A prototype MCP server on your real systems typically takes two weeks. Production hardening and governance depend on the number of systems and your compliance requirements.
Ready to connect AI to your platforms?
Start with a two-week prototype on your real systems, then scale it with the security and governance your organization needs.