Concepts
Averment provides a structured way to control AI actions and define how your system behaves.
Decision
A decision is any action your system is about to take that could have consequences. Such as sending money, calling an external API, updating a database, or running a workflow.
You describe the intent of that decision in natural language, with optional structured context.
js
Decision signal
The decision signal is the output of Averment. It is derived from multiple layers of intelligence.
It contains:
- riskLevel - How risky the action is (
low,medium,high,critical) - summary - A short human-readable explanation
- action - What must be enforced (
allow,caution,review,block) - attentionRequired - Whether human or system attention is needed
- reasons - Clear explanation of why the decision was made
- engineVersion - Which engine produced the result
Enforcement action
Enforcement is how Averment turns a signal into a concrete outcome.
| Action | Meaning |
|---|---|
| allow | Execute immediately |
| caution | Proceed with awareness; soft gate |
| review | Pause and require active attention |
| block | Stop execution |
Averment always returns a final action, even if input is incomplete. If the system cannot determine a clear action, fallback logic applies:
critical→blockhigh→reviewmedium→cautionorallowlow→allow
riskLevel and action are related but not the same. Risk level informs the action, but rules and context can override it.
Next steps
Now that you understand the core concepts, learn about the engines that power Averment's decisions.
Engines
Understand what engines are, how they work, and how to choose one for your use case.