Capabilities
NeverBounce gives an AI agent a disciplined way to discover likely business addresses, verify candidate mailboxes, preserve uncertainty, and prepare a human-reviewed handoff. The workflow does not turn a database response into permission to send.
Start with a person’s professional name and a confirmed company domain. The agent can evaluate common address patterns and provider evidence, then label the output as a candidate until verification completes. Similar names, subsidiaries, aliases, redirected domains, and recent company changes remain review conditions rather than being silently merged.
Inspect: domain identity, role relevance, pattern evidence, checked time, and unresolved alternatives.
Syntax checks remove malformed addresses; domain and MX checks establish whether mail routing exists; provider-level responses can add evidence about mailbox status. Results remain separated as valid, invalid, catch-all, or unknown. Temporary network failures, throttling, greylisting, and provider policies can limit certainty, so an unknown state is a useful result rather than an error to conceal.
Inspect: state definition, provider response, retry policy, and freshness window.
Verification should happen before an address enters an outreach queue. Invalid results are excluded; catch-all and unknown results follow a documented risk policy; prior opt-outs and suppression entries override a fresh technical result. Duplicate records are resolved before export so the same professional contact is not contacted through parallel campaigns.
Inspect: suppression source, deduplication key, owner, and approval status.
A technically valid mailbox does not guarantee inbox placement. Operators still need healthy SPF, DKIM, and DMARC configuration, stable sending behavior, relevant copy, opt-out handling, and monitoring for bounces and complaints. NeverBounce keeps address verification separate from message drafting and sending so a named reviewer can challenge relevance and local communication requirements.
Inspect: authentication, sending policy, regional basis, and human approver.
Include known valid, invalid, catch-all, disposable, role-based, and ambiguous examples. Record expected state before the run, then review disagreements without collapsing them into a universal accuracy claim.
npx -y @okki-global/okki-go-taroball