PII detection
Personal data guardrails catch names, identifiers and contact details before they reach the wrong tool or output.
Governance
AI Agent Governance
Every agent runs under a governed execution contract. No action happens without guardrail checks, write gates and an audit entry. The rules sit in the platform so no single agent can skip them.
The runtime contract
The governed execution contract
Five steps run in the same order on every agent action. Your risk team reviews one contract instead of a different design per agent.
Model call
The agent asks a model for its next step through the policy-driven model router under your tenant configuration.
Guardrail assertions
The platform checks the response before anything happens: PII rules, contact rules, safety predicates and more.
Tool calls against an allow-list
Only the tools listed in the agent profile may run. Anything not explicitly marked as a read is gated as a write.
Write gate
Any action that changes a record waits for a named person to approve, modify, take over or reject with a reason.
Audit append
The step, its checks and any decision land on a hash-chained, tenant-scoped ledger that anyone can verify.
Guardrails
Guardrails
Over 100 predicate packs check every model response before it reaches a tool or a person.
Collections FDCPA and TCPA
Contact frequency limits, do-not-call lists, dispute recognition, hardship detection and consent rules enforced on every step.
Safety predicates
Content safety, self-improvement blocks, honesty validation and data integrity checks run after each model generation.
Tool scope enforcement
Default-deny on tools. Anything not explicitly listed in the agent profile is refused and logged. Unknown tools are gated as writes.
Autonomy levels
Autonomy levels
Autonomy is earned through evals, not granted. Every agent starts at shadow and earns its way to autonomous through measurable performance.
Shadow
The agent runs but its actions are recorded, not executed. Your team reviews what it would have done.
Assisted
The agent proposes actions. A named person decides: approve, modify, take over or reject with a reason on record.
Autonomous
The agent acts within its guardrails. Exceptions still escalate. Autonomy is earned through evals, not granted by default.
Safety
Kill switches
Circuit breakers and kill switches that halt any agent instantly. When something goes wrong, the platform stops the agent before another action runs.
Per-agent kill switch
Halt a single agent immediately. Every action in progress is stopped and the switch is recorded on the audit ledger.
Per-tenant circuit breaker
Pause all agents for one tenant at once. Useful when an incident affects one customer and you need everything to stop.
Fleet-wide pause
Pause every agent across every tenant in one action. The platform resumes only when a named operator lifts the pause.
Govern your agents from day one.
Book a demo and we will show the execution contract, the guardrails and the kill switches running on a real agent.