Governance

Built so you always know what your AI employee is doing

Every action your AI GTM employee takes is logged with a cost and an owner before it fires. You can review sends, external writes, and verifications before they happen, and spend ceilings cap every paid provider. These are not settings you configure later: they are the architecture.

How the controls work

Every metered action is logged before it fires

Every email draft, contact verification, external write, and LLM call emits an audit event carrying the use case, action kind, estimated cost, and a timestamp. The ledger is append-only and exportable, so you can always see what your AI employee did and what it cost.

You can review anything before it goes out

Sends, external API writes, and verification calls queue in the console with the reasoning and cost attached, so you can check them before they happen. Your AI employee proposes; you can look at the work as closely as you want, and dial that down as it earns your trust.

Spend gates and a kill-switch on every paid provider

Every adapter that calls a paid external service carries a per-run budget ceiling and a master kill-switch. If the ceiling is hit or the switch is off, the call does not happen. Stub mode lets the full workflow run end-to-end during development without burning a single credit.

Sourced contacts carry a provenance record

Every contact record stores the date it was sourced and the source it came from. We never ship a contact without a dated, traceable origin. If the source is older than six months, the record is flagged for re-verification before it is used.

One schema per client, never shared tables

Each client's data lives in a dedicated schema. There are no shared tables between clients and no tenant-id columns that a query could accidentally omit. Isolation is structural, not a filter.

Ownership and visibility are permanent architecture

The audit log, the spend gates, and the review queue are not training wheels for an early product; they are the architecture your AI employee runs on. As it earns more autonomy over time, you keep the same visibility into what it does and why, and the same ability to step back in whenever you want.

Sample audit-event ledger

Every metered action produces a row. The ledger is exportable. Values below are illustrative fixture data, not live records.

Action kindUse caseTargetEst. costTimestamp (UTC)Status
email_draftcold-outreachacme-example / intro sequence$0.0032026-07-28 09:14 UTCPending approval
email_sendcold-outreachacme-example / intro sequence$0.0012026-07-28 09:22 UTCApproved
contact_verifysignal-outboundriverdale-co / 3 contacts$0.0302026-07-28 10:01 UTCExecuted
icp_publishicp-targetingriverdale-co / ICP v3$0.0122026-07-28 10:45 UTCApproved
email_sendcold-outreachnorthpeak-inc / follow-up$0.0012026-07-28 11:03 UTCRejected
content_draftseo-contentblog / answer-engine post$0.0182026-07-28 11:30 UTCPending approval

We never ship a contact without a dated source

Every contact record stores the date it was sourced and the channel it came from. If a record is older than six months it is flagged before it is used. The examples below show the provenance fields that travel with every contact.

Name (example)RoleSourceDate sourcedVerified
A. Rivera (example)Head of GrowthPublic LinkedIn profile2026-07-25Verified
B. Chen (example)Founder & CEOCompany website2026-07-24Verified
C. Okafor (example)VP MarketingPublic LinkedIn profile2026-07-20Pending

Controls you can audit, not just read about

The audit ledger, review queue, and spend gates are live in the product today. Request access to see them in the console.

Learn how Natively works →