I built a multiverse for AI agents: they rehearse every action in hundreds of graph copies before touching anything

Description and use

Multiverse lets AI agents rehearse each action in hundreds of graph copies, measuring the chance of critical failures and blocking unsafe plans.

Problem

AI agents perform actions on real systems, but silent dependencies can cause critical failures when an action is taken.

Outcome

Multiverse blocks unsafe actions when the probability of breaking a critical component exceeds 5%, reducing risk of silent failures.

Workflow

• Snapshot the company graph with GRAPH.COPY
• Create a fresh copy for each simulated future
• Roll dice for every uncertain link to decide if it exists in that future
• Apply the planned change in the copy
• Run a ripple query to detect failures and record what broke
• Delete the copy

Evidence

AI agents are starting to act on real systems. They clean up databases, rotate keys, move cron jobs, scale services. The action itself is rarely the dangerous part. The danger is everything that silently depends on the thing being changed.

Source

Multiverse treats the company as a graph in FalkorDB, and every uncertain link as a coin with a known bias.

Source

Each future becomes one square on screen. Orange means something business-critical broke. After enough futures, Multiverse knows the chance of breaking something critical, and if it's 5% or more, the plan is blocked.

Source