ABOUT CIRCUIT
Authorization is a moment.
Autonomy unfolds over time.
CIRCUIT began with a question: after an AI agent receives permission to act, how do we keep verifying that its evolving behavior still matches what the user actually authorized?
THE STORY
Built for a world where agents do more than answer.
Created for the Binance Agent OS Mini Hackathon, CIRCUIT started from the gap between static permission and continuous behavior. An agent can remain technically authorized while its retries, frequency, evidence, market assumptions, or product choices drift away from the user’s original intent.
The answer was not another prediction model. It was runtime supervision: a Financial Mandate, deterministic policy veto, sequence-aware drift detection, sticky breaker states, and a tamper-evident Flight Recorder.
THE THESIS
Permission is necessary. Continuous verification is different.
CIRCUIT treats autonomy as a process that must stay inside an operating envelope—not a one-time authorization event.
THE NAME
CIRCUIT
Continuous Intent & Runtime Control for User-authorized Intelligent Transactions.
The name reflects the closed-loop idea: observe intent, evaluate policy, detect drift, control execution, record evidence, repeat.
DESIGN PRINCIPLES
Deterministic vetoAI can interpret and explain; code owns the financial boundary.Explicit activationGenerated policy is draft until the user deliberately activates it.Sequence awarenessSafe-looking individual actions can form unsafe behavior over time.Evidence before actionStale or uncertain state is a reason to stop, not improvise.AuditabilityEvery verdict should be replayable and inspectable after the fact.No trading-performance promiseCIRCUIT is a control plane, not a guarantee of profit.
SEE IT RUN
Attack the control plane yourself.
Use the Simulation Lab to reproduce eight containment scenarios, then inspect Mission Control to see the runtime state and trace evidence.
