Marketing Tool for In-House AI Agents
AI agents are in demand - and bring their own security risks the moment they dock onto foreign systems via APIs. YourValidator deliberately takes the other path: it offers no agent interface. Whoever runs their own agent needs only the JSON backup file. The agent prepares the numbers, the tool computes and visualizes - without the two ever being directly connected.
Why an API would be the wrong promise here
Every interface is an attack surface
An open API sounds convenient, but opens exactly the door that data-sensitive organizations want kept shut: live access to the tool and potentially to a server. For IT security, that's often the point where a tool drops out.
What plain file handover changes
Without an API there is nothing that could be addressed from outside. The only touchpoint is a file you generate yourself and import yourself. Data acquisition and agent logic stay entirely in your environment - the tool remains a pure compute engine with no attack surface.
How the compute-engine approach works
- Run your own agent. The agent runs in your environment - not in the tool, not on a server of ours.
- Bring the real numbers together. The agent pulls values from your sources - Meta and Google ad accounts, CRM, licensed market data if needed - and computes metrics and probabilities per channel.
- Write into the JSON file. The agent stores the results in a file that matches the structure of the JSON backup.
- Import via the existing function. Load the file via the existing import function - no special route, the same function as for any backup.
- Compute and visualize. YourValidator takes over calculation and display - S-curve, saturation, halo, break-even, CLV - on the numbers the agent delivered.
Export a JSON backup and inspect the structure: try the tool
Prerequisites & limits
What you bring
- Your own AI agent and the data connection - both in your environment, outside the tool.
- The mapping of your values into the structure of the JSON backup; a backup exported from the tool serves as the template.
- Responsibility for data acquisition, rights to market data and data protection - it stays with you.
Where the limit lies
- There is no API, no live connection and no access to the tool or a server. The only touchpoint is the JSON file.
- This is an open possibility, not a finished product feature - and no promise of a stable interface format. The backup structure can change; you maintain the mapping yourself.
Common thinking traps
- Expecting a programmatic interface. There is no API and no live format. Whoever wants to build a fixed integration builds it against a file that can change - not against a guaranteed contract.
- Mistaking the compute layer for the data layer. The tool doesn't check the delivered numbers for correctness. Whatever the agent merges wrongly, the model computes cleanly wrong onward.
- Seeing the agent as part of the tool. Data acquisition, agent logic and their security lie entirely with you - the tool bears no responsibility for them and offers no function for them.
Frequently asked questions about the compute-engine approach
Is there an API to connect YourValidator?
No. There is no API, no live connection and no programmatic access to the tool or a server. The only touchpoint is the JSON backup file, which you load via the import function.
How does my agent get the right file format?
Export a backup from the tool once - it serves as a structure template. Your agent writes its results into the same structure, which is then loaded via the import.
Where does the data acquisition run?
Entirely in your environment. The agent brings your sources together - ad accounts, CRM, licensed market data if needed - and computes the values from them. The tool sees nothing of this except the finished file.
Is this an official product feature?
No. It is an open possibility that follows from the local architecture - not a finished feature and no promise of a stable interface format. The backup format can change; you maintain the mapping yourself.
Why doesn't YourValidator offer an agent itself?
For security reasons. An agent with live access would be an attack surface that contradicts the local, serverless approach. The separation - your agent, our compute layer, only a file in between - deliberately keeps the tool clear of it.
