Debug why your AI agent
changed behavior.
Decision DNA records the experiences behind agent decisions, replays suspicious decisions under controlled interventions, and measures which memories, sources and beliefs actually changed the outcome.
“Logs tell you what happened. Decision DNA tests what shaped the behavior.”
Counterfactual Decision Lab
Edit the agent’s memories and priorities, then replay the decision with each influence removed. Everything executes locally in your browser.
Procurement Agent
Choose the supplier that best balances reliability, quality and cost.
D-0001
Awaiting replay
Run the agent once, then Decision DNA will replay the choice with each memory removed.
Trace a bad decision back to experience.
Decision DNA links observations, memories, beliefs, plans and outcomes so Guardian can reason over an agent’s behavioral history without trusting the agent’s self-explanation.
The debugging layer beneath Guardian.
Decision DNA separates stated_reason from measured_influence. Guardian sits above the records, detects behavioral drift, ranks suspicious influences, plans replays and proposes the smallest safe repair.
{
"agent": "Procurement-3",
"symptom": "supplier preference drift",
"suspect": "M-144",
"origin": "shipping-api:event-881",
"measured_influence": "high",
"repair": {
"action": "downweight memory",
"requires_approval": true,
"verify_with_replay": true
}
}Guardian: the debugger
for autonomous agents.
The next real milestone instruments a persistent agent, watches its behavior over time, detects drift, traces a suspicious decision back through memory and source lineage, and tests a proposed repair in replay before touching production.