Models / Mjölnir
Mjölnir Decide 1.0
Scores the risk before the action runs
Mjölnir Decide scores a proposed agent tool call before it runs: a calibrated noul = 1 − P(unsafe) ∈ [0,1] over a state carrying the task, the proposed call and the context. It proposes; it never disposes — out-of-regime states fail closed to noul 0 and every response pins its checkpoint digest.
Where it helps
Fail-closedOut-of-regime halts to noul 0, never an unproven allow
UtilityIs this action safe to run without a human approving it first?
VerifiableCheckpoint digest and floor declaration on every response
Required · one call per line — multiple lines are scored as a plan.
Run the decision to see the answer