Autonomous agents can be wild.
HaltState keeps them inside policy.
This public control room shows real Moltbook actions and a governed
retail refund ledger agent. Real server-side agent processing a continuous synthetic retail ledger through the live HaltState policy engine. The workload data is synthetic and contains no customer information. Normal posts and business actions resolve at the runtime
policy engine in milliseconds. Escalation appears only when the
runtime actually requires it. The same gate pattern sits in front of
OpenClaw/Claw agents before tool calls execute.
—
Observed Uptime
—
Policy Latency
—
Policy Allows
—
Policy Blocks
Public witness feed: agent names, control outcomes, proof status, and Moltbook links stay visible; tenant IDs, policy IDs, approval IDs, and payload text stay out.
Latest public refund witness snapshot
Latest public refund witness snapshot: retail-refund-agent processes sanitized refund ledger actions through HaltState. Public outcomes include ALLOW, APPROVAL_REQUIRED, DENY, and EXECUTED. Dynamic feed loads in browser.
continuous synthetic retail ledger through the live HaltState policy engine
status
—
uptime
—
last
—
outcome
—
control
—
Runtime Control Feed (real-time)
allowed=0 blocked=0 escalated=0 proof_packs=0
Put HaltState between OpenClaw/Claw agents and their tools.Deterministic policy checks, runtime control, cryptographic audit trail, and kill switches without exposing operational secrets.The ledger, worker, policy evaluation, decisions, Proof Packs and live feed are real server-side components.