Kiwi

GTM teams such as RevOps, BizOps, Sales and Marketing rely on manual, repetitive workflows like writing renewal account summaries, reviewing pipeline, or turning call notes into tickets. Today these tasks are done by hand across scattered tools like Gmail, Salesforce, Gong, and Slack, with no shared system of record for the decisions made along the way. Teams end up maintaining duplicate or stale skill files and workflows, and have to file data infrastructure requests to get the information they need. Without traceability back to the underlying data, AI-generated outputs are hard to trust, especially when source data is out of date or incomplete.

Kiwi lets a user describe a workflow in plain language, then drafts the agent steps, tool integrations, and human approval checkpoints needed to run it, pulling data directly from tools like Salesforce, Gong, Gmail, Slack, Calendar, Notion, HubSpot, Linear, and Zoom. Every claim an agent makes is cited to a synced data source, with snapshots and SQL queries versioned so runs can be replayed bit-identically and discrepancies traced to specific changed rows. If a required data source hasn't synced recently enough, the run stops rather than returning stale or incomplete results. The system maintains a skill library automatically, merging duplicate workflows and flagging stale ones, routes approvals to team members, and uses feedback from each run to improve the agent over time. Example use cases described include scanning Gong calls for expansion signals, building account timelines into pilot plans, and turning Granola call notes into Linear tickets.