Architecture

Amfelt is organized in five planes. The diagram below shows all of them; the sections that follow walk through each.

User Plane  

People sign in through the web application. Machines drive the CLI and the REST API. A governed Model Context Protocol (MCP) endpoint for AI agents is on the roadmap (shown in the diagram).

Control Plane  

The control plane is where configuration lives and orchestration happens: the application, the API, data contracts, the AI advisor that recommends a table type for each object, plus the platform’s own database, cache, and Git server. It is built to run as a managed service on Scaleway in Amsterdam, behind Cloudflare, and production is not live yet; Security & Trust says where each leg runs. If you need a private-cloud or on-premise control plane, contact us.

Security Plane  

Sign-in runs on a separate identity service, Hanko, and secrets live in a separate vault with its own database, which the API reaches over the platform’s private network (until the first customer, the vault shares the platform’s cache). Authorization runs inside the API. The Security Model covers this plane in depth.

Observability Plane  

The agent reports each run’s progress to the control plane as it goes, and that is what the app’s Monitor pages show you: run status, checkpoints, and metrics across runs. The agent’s own log stays on the host it runs on. Exporting telemetry to your own observability stack is not available.

Data Plane  

The agent and its connectors, your sources, and your data platform. This plane lives inside your infrastructure: data is extracted and loaded there, and transformed by SQL that runs in your data platform. Configuration, contracts, run results, connection credentials, connection-test answers, and advisor statistics cross the boundary, as Security & Trust sets out. Deployment describes the options.

For the evaluator-level summary of this architecture, see How It Works .