AI is powerful. Every interaction deserves validation.
Validate Inputs. Protect Data. Govern AI Actions.
An API-first platform that checks AI inputs and outputs against configurable policies: before a message reaches your model, and before an answer reaches your user.
Watch GuardOath in 2 minutes
An AI narrator explains GuardOath’s approach to input and output validation and configurable policies. Development preview: some capabilities shown are planned or in progress.
See what your AI checks catch.
Follow the conversation: enter a message, check it, paste your AI’s answer, then check the answer before releasing it.
Connect your application
Your API key stays in this tab’s memory and is sent only to this server.
Demo users run local checks only and stop at a monthly limit. Visitors who pick the same demo user share its settings and limit.
Checks on this record
Each check is a field on the tenant record. Choose what happens when a check finds something, or turn it off. Select What is this? under any check for what it looks for and why you would use it. Changes apply from the next check. The checks shown follow the check profile selected under Test a conversation: the record’s own checks can be changed here, and an example profile’s checks are shown as they run.
- Off
- The check does not run.
- Block
- A finding stops the message or answer.
- Flag
- A finding is reported and marked for review; your application decides what to do.
- Allow
- A finding is reported, and the message or answer goes through.
- Redact
- Personal data only: what is found is replaced, and the checks run again on the result.
Test a conversation
Local checksFour steps: check the user’s message, get your AI’s answer, check the answer, then release it. Every step below belongs to this conversation test.
Advanced: application rules and focused examples
A policy is a saved bundle of checks and decisions. Normal conversations use the combined profile automatically. Change it only to explore a focused example or application-specific rules.
Connect to load the combined conversation profile.
1 Input data
Input examples replace only the user message and its format. The proposed answer and check profile stay unchanged. Examples follow the active profile; choose another profile under Advanced to try its examples.
2 Input test
Check the message before sending it to an AI model. Findings, incomplete checks or simulation hold the next step, even in observe mode.
Input checks
Waiting for a check.
3 Output (AI) data
Complete the input test first.
In your application, an accepted input goes to your AI model here. This local demo makes no model call. The proposed answer box stays empty and locked until the input test passes; then paste a proposed answer or use an output example. If the input is held, there is no proposed answer: change the user message and run the input checks again.
Output examples replace only the proposed answer and its format. Your user message and completed input test stay unchanged.
4 Output test
Check the proposed answer before it reaches your user.
Output checks
Waiting for the input test.
Run a prepared example through the sequence
Prepared example mode: you may choose the proposed answer before the input test. Input still runs first, and the answer is checked only if input passes. If the proposed answer is empty, a sample safe reply is used.
Conversation status
Ready to testConnect and run the input test to begin.
Answer preview
Not checkedAn answer appears here when the configured checks complete without findings.
How an application uses GuardOath
Check the user’s input → call your AI model → check its answer → apply your policy’s decision.
The same input and output API endpoints power this playground. There is no separate demo scoring system.