FlyBy: Agentic AI for Content Systems
On large websites, information is spread across many pages and different sources. Further knowledge sits with individual staff members and can be lost when people move on. A migration requires content to be reviewed, consolidated and adapted to the new page structure.
Panter has developed FlyBy, a system for migrating content inventories, consolidating content and creating new pages from existing information on the basis of a briefing. Specialised AI agents research the inventory and build pages from it with the components of the design system. An agentic workflow keeps information and its context in a traceable structure over time. The content can then be edited directly in FlyBy or handed over to a CMS.
Connect
Chat for the editorial team, a REST API for automated jobs and MCP for in-house AI agents: three entry points lead to the same engine.
Create
Existing pages, scattered content or briefings become pages built from the components of the design system.
Hand over
The editorial team edits content directly in FlyBy. For export to a connected CMS, the system validates the output against that system’s schema.
Connecting systems
In the chat interface, the editorial team works in dialogue with the AI agents. Existing systems submit their jobs automatically through the REST API. In-house AI agents reach the tools of FlyBy through the MCP server, an interface for AI applications.
FlyBy can be used on its own or connected to existing systems. Editorial work in the chat and automated jobs run on the same engine. The entry points can be combined depending on the task.
Creating pages
One existing page becomes a revised page. Several scattered pages become one consolidated page. A briefing becomes a new page.
The agents research the existing content inventory and build pages from it with the components of the design system. A semantic index opens up information that fits the subject. The knowledge base holds the specifications for page types, content structure, language and terminology.
Handing over content
The editorial team can view and edit the created content directly in FlyBy. For a connected CMS, FlyBy delivers the content as JSON. Fields and components follow the agreed target structure.
The editorial workflow determines how much review takes place before export. That can be sign-off on every single page in FlyBy, or a downstream check in the CMS. When content goes live follows the existing publishing process.
Matched to the CMS
The schema of the connected CMS defines the expected fields and components. FlyBy validates the output against this schema at handover. Every export refers to a specific schema version.
Large jobs run in the background. The connected system polls the processing status and collects the results. We agree the integration with the business team and IT.
A CMS migration, the revision of a content inventory or a new content solution can be the starting point. We set FlyBy up for standalone use or for the integration with existing systems. Together with the business team and IT, we define the tasks of the agents. That includes which knowledge they need and how the results are processed further.
During setup, we align instructions, knowledge base, components and technical checks with one another. The output of language models can vary. Content therefore has to be open to editorial review and to a check of its technical structure.
Tools for content work
FlyBy connects work on content with research in the inventory, editorial specifications and interfaces. The following building blocks work together.
Six characteristics of FlyBy
- Model choice: The language models are interchangeable. Whether a model is suitable can be assessed against the content and the tasks at hand.
- Agent integration: In-house AI agents can call the tools of FlyBy through MCP and include them in existing workflows.
- Target systems: The architecture provides for further CMS integrations. The specific integration depends on the target structure and the available interface.
- Review process: Content can be edited in FlyBy or in the connected CMS. An editorial sign-off can take place before export.
- Consistency: The shared component library and the specifications of the knowledge base form the basis for the language and the structure of the pages.
- Content volume: Automated processing can also be applied to the many rarely visited pages of a large content inventory.
At Swiss Post
10,000 pages, four languages
Around 10,000 corporate pages in four languages and 5,500 images make up the inventory in the CMS migration of Swiss Post. 80% of visits go to 20 pages. The many less frequently visited pages have to be processed as well.
The estate has been crawled and vectorised. The pages sit in the database. The AI agents use them as a research source and build pages in the agreed target structure. Panter developed FlyBy for this together with Swiss Post. Swiss Post fetches the results through the interface and takes them into its CMS.
Risks and limits
- Content quality: The validation against the CMS schema checks the technical structure. Factual errors remain possible. Knowledge base and research support the creation; the editorial team can review and edit the content in FlyBy or in the CMS.
- Processing state: With many parallel jobs, the state of processing has to remain traceable. Job status and review status show the progress. The steps for editing, review and export are defined for the workflow in question.
- Model providers: The language models are connected through a gateway and are interchangeable. Before a switch, we assess how the new model affects the quality of the content and the behaviour of the agents in the intended tasks.
- CMS integration: A connected CMS is not a prerequisite. To integrate a CMS, we clarify the target structure and the interface with the business team and IT. A CMS decision that is still open is taken into account.
- Data and governance: Before deployment, we clarify which content may be processed and which policies apply. A first run can be limited to suitable publicly accessible content.
- Brand consistency: Language and structure can differ from one creation to the next. Component library and knowledge base provide shared specifications. The editorial team can review the content against these specifications and revise it.
Michele Romano, Senior AI Architect & Consultant
Interested in our solution?
Talk to Michele about how you want to create, structure and develop your content. Together we clarify how agentic AI can support your workflows and which technical implementation fits your plans.
