Agents
An Agent is any system that can:- Reason
- Decide
- Invoke tools
Tools
A Tool is any capability an agent can invoke:- API calls
- Database queries
- File access
- Emails, messages, jobs, side effects
- What kind of data they touch
- Who they affect
- Whether they’re reversible
- Where outputs go
Tool Calls
A Tool Call is a specific invocation of a tool during a run. It includes:- Inputs
- Context
- Timing
- Outcome (if allowed)
Rules
A Rule is a deterministic check evaluated against context. Given the same inputs, a rule will always return the same result. Rules can inspect:- Agent identity
- Run context
- Tool metadata
- Tool call parameters
- Historical signals
Actions
An Action is the consequence of a rule evaluation. Actions are hard, not advisory. Examples:- Allow
- Block
- Modify inputs
- Require approval
- Kill a run
- Lockdown
Determinism
The same inputs produce the same decisions. This enables:- Replay
- Debugging
- Audits
- Trust
Why This Matters
In regulated environments, - “this is unlikely to happen” - isn’t good enough. You need to be able to say:- This agent could not access this data under these conditions
- This action was blocked for this reason
- Here is the exact rule version that made that decision