Agent overview
An agent is how a read becomes an action without you at the keyboard. You build a scoped agent that watches the signal and executes only inside the hard limits you set. It acts when your conditions hit, and never beyond what you allow.
This is the most powerful surface in intenthybrid AI, and the one that is designed most carefully, because it touches funds. The guiding principle is simple: an agent does exactly what you wrote, and nothing else.
What an agent is made of
Every agent is defined by a trigger and an action, wrapped in limits.
- Trigger. The condition that wakes the agent, for example crowding crossing a level on a market.
- Action. What it does when the trigger fires, for example reducing an exposed position toward USDC.
- Limits. The boundaries it can never cross, covered in detail on Guardrails and lifecycle.
An illustrative agent
Consider a "guardian" agent configured like this:
- Trigger: crowding above 80 percent on the chosen market.
- Action: reduce to USDC.
- Max per trade: 500 dollars.
- Slippage cap: 1.0 percent.
- Expiry: 24 hours.
- Auto-execute: on.
- Kill switch: armed.
The values are an example. The structure is the contract: a clear condition, a clear action, and a fence of limits around it.
Why scoped, not autonomous
intenthybrid AI agents are deliberately not open-ended. They do not roam, optimize, or make judgment calls. They watch one signal, take one defined action, and stop at the edges you set. That narrowness is the feature. It is what makes it reasonable to let software act on a read while you are asleep.
Where agents act
An agent executes onchain from a wallet you connect, which is why connecting a wallet is part of arming one. Execution is bounded by the guardrails on every call. See Guardrails and lifecycle for how the limits are enforced and how to revoke an agent instantly.
Status
The create-agent flow is in active development. See the Roadmap for current status.