Open Data Agent
An agent that queries open data without paying for it
We designed the architecture for a Claude agent to query public data (mobility, air quality, safety) in real time, without the server hosting the connection ever seeing or paying for a single model token: the reasoning always lives in the assistant's own account, and the server only translates the request to the public source and returns the JSON.
Not a client case: a proof of technical range for the systems the studio can build.
Who pays for the intelligence?
Putting an agent on top of public data usually means a server that calls a model and pays for every token, whoever asks. The infrastructure bill and the intelligence bill end up on the same side.
Two bills that never cross
Reasoning in the user's account
Claude reasons and pays its own tokens in the account of whoever uses it.
A server that only translates
The server turns each tool call into an HTTP request to the public source and returns the JSON. It never sees a model token.
Nine tools, free sources
Nine tools are exposed to the agent, and the public sources it queries cost nothing.
The path of a question
Claude
Your account
TB's server
Translates only
Public source
Mobility · air quality · safety
Reasons. Pays its own tokens.
$0. Never touches a model token.
$0. Always open.
Two bills that never cross: the agent’s own account pays for the reasoning; the server never sees a model token.
Need a system like this one?
We design it, build it and leave it running in production, inside what your company already operates.