AI agents // SINT Labs
SINT Protocol
Checks what an AI agent is allowed to do before it acts.
- Role
- Founder · Protocol and product design
- Period
- SINT Labs, 2024–; working paper, 2026
- Contribution
- Authorization model and product direction
- Result
- Open-source code and a working paper
The problem
An agent can propose an action without having the authority to carry it out. The gap matters when a system can spend money, delegate work, use tools, or interact with the physical world.
SINT explores how to check that authority at the point of execution, rather than relying only on what an agent was told in a prompt.
My contribution & approach
The central design decision is to separate proposing an action from authorizing it.
The design makes three questions explicit: what can this agent do, how much can it spend, and when must a person decide? Scoped permissions, limits that carry through delegation, and authorization records turn those questions into product controls.
Collaboration & constraints
The implementation, contribution history, and working paper are public so the design can be inspected and challenged.
Integrations need to enforce the check before execution. A protocol cannot make an arbitrary tool or robot safe by itself: enforcement, configuration, and the surrounding system still matter.
What is available
The code and paper are available for inspection. This is an evolving open-source project, not a certification or a promise that all agent behavior is safe.
Explore the code ↗ Read the working paper ↗ Try the simplified interaction →