Ballet

Ballet targets the teams that spend their days moving data between CRM, support, billing and communication tools. Instead of mapping fields by hand or buying a separate connector for every system, an operator writes a short description of the process — for example, drafting a closed-won handover from the last sales call and posting it to the account owner. Ballet turns that description into a multi-step playbook with triggers, integration calls, AI drafting steps and human approval gates.

The platform supports connectors for services such as Salesforce, HubSpot, Gong, Slack, Zendesk and Notion, and can also call MCP servers or any API. When a service has no existing connector, Ballet writes the integration as part of the build, which removes one of the biggest blockers for automation programmes. Runs can be triggered on demand, on a schedule or from a webhook, and can also be invoked from the Chrome extension, the Claude connector or the Slack agent so work starts where the user already is.

A distinguishing design choice is the split between fixed rules and AI. Steps that must be exact run as deterministic code, while model-driven steps are reserved for judgement-heavy work like summarising a call or drafting a reply. Every run is logged with what each step read, what it changed, and how long it took, which gives operations teams an audit trail they can trust. Ballet also includes a reporting layer that answers ad-hoc customer questions across multiple systems without a separate BI project.