Constantly, dogfooding is a stated discipline. Falkster's own knowledge runs through a production Heidi install, the agents that watch Falkster's metrics are Heidi agents, and the growth workflows are built on her own API and MCP surface.
A company selling a company brain that did not run on one would be a red flag, so the commitment runs the other way: Heidi is Falkster's first and most demanding customer, and rough edges get found at home first.
The machine surfaces are part of it: the agent card, the MCP endpoint, and the llms.txt files on falkster.ai are the same machinery every customer install publishes, in production, in use.