Infrastructure decision tool
Which AI Agent Architecture Should You Use?
Five questions decide most agent infrastructure arguments: what runs the model, where the data may go, when the agent must be reachable, how much memory it needs, and what recurring spend is available. Answer them and this tool returns one architecture, the reasoning behind it, and the answers that cannot all be satisfied at once.
How the recommendation is made
The answers are applied in order of how decisive they are. A hard requirement to keep data on hardware you control is checked first, because nothing rented can satisfy it. After that, whether the machine serves a model at all separates an orchestration host from an inference host, and only then do uptime, memory and budget narrow the choice.
Where your answers cannot all be satisfied — a hosted model API alongside an on-premise requirement, or rented capacity with no recurring budget — the tool says so rather than quietly dropping one of them. Every one of the 432 possible answer paths is covered by tests.
Use this architecture finder on your site
Embed the deterministic five-question finder free with attribution back to its methodology and full decision context.
Put numbers behind the recommendation
The finder chooses a shape. The two calculators price and size it: the memory question above is answered properly by the RAM and VRAM calculator, and the recurring-budget question by the hosting cost calculator.
Agent Infrastructure Recommendation Kit
Turn the recommendation into a defensible decision record
Five questions narrow the architecture. The Recommendation Kit carries it through requirements discovery, capacity planning, option comparison, evidence capture, validation, and a client-ready deliverable.
Get the Recommendation Kit — $19