AI that stays inside the lines.
Guardrail policies check what goes into and comes out of your AI agents - spotting personal data, blocking words and patterns you never want to see, and keeping a record of every violation. AI spend is capped before it runs.
Rules you write. Checks on every run.
Policies are simple to set up: pick what to look for, where to look, and what should happen when it is found.
Personal data detection
Spots email addresses, phone numbers, card numbers, national IDs (such as Aadhaar, PAN and SSN) and IP addresses in agent input or output.
Keywords and patterns
Block or flag the terms and regular-expression patterns you define - internal code names, competitor names, account formats.
Block, warn or log
Choose per policy: stop the run, let it continue with a warning, or simply record the match.
Before and after the model
Check the input, the output, or both. A blocked input stops the run; a blocked output is withheld from the user.
Scoped to an agent
Apply a policy to a specific agent, so a payroll agent and a marketing agent can follow different rules.
Violation log
Every violation is recorded against the run that caused it, with personal data masked in the log itself.
AI Credits caps
An organisation-wide credit pool and per-user allocations. Agent and Studio runs reserve credits up front and pause instead of overspending.
Usage alerts
Get alerted when usage crosses a share of your plan, when spend spikes, or when runs get stuck or start failing.
Three steps. No developer needed.
Write a policy
Choose the check (personal data, keyword or pattern), where it runs (input, output or both) and the action.
Agents are checked
Each agent run is checked against the policy before the model sees the input and before the user sees the output.
Review violations
Open Guardrails to see what was blocked, warned or logged, and tune the policy.
Put AI to work - on your terms.
Guardrails sit alongside human review and the audit trail, so every AI action is checked, approved where it matters, and on the record.