product

missionsagent-driven promotion campaigns with adjustable autonomycontent & draftsAI-generated content reviewed before it shipsinbox & engagementsunified inbox for mentions, drafts, and AI decisionsCRM & contactsrelationship management with AI-powered outreachresearch & recordsstructured data gathering on autopilotworkflowsvisual automation graphs — no code requiredprojects & docswork planning, docs, and standups — all native

solutions

for startupsyour marketing team is an agentfor agenciesscale clients, not headcountfor creatorsgrow your audience while you createfor enterpriseautonomous marketing with full controlview all solutions
platformmodelspricing>docs

Using the AI

The agent's toolbox is built from the same action registry that powers the API. Every action becomes one tool; the agent selects tools by name and the AI SDK validates arguments with the action's own schema before the handler runs. One source of truth — see Core Concepts.

How a tool is named and described

  • Name: the action name with dots → underscores. missions.create becomes the tool missions_create.
  • Description: the action's description, plus any llmHints appended as "When to call" guidance, plus a note when the action is mutating.

The hints are how you steer the agent toward the right tool. The Reference shows each tool's name, description, and hints.

Mutating actions & autonomy

A mutating tool changes platform state. How that's handled depends on the agent's autonomy level:

  • Autonomous agents execute mutating tools directly.
  • Supervised agents queue a Decision instead, for a human to approve or reject before the change lands.

After a mutating tool runs, the result carries a refresh hint so any open UI reflects the change without a manual reload.

Errors are real failures

When a tool's handler throws, the agent sees a genuine failure (not a success containing an error field). This stops the agent from marching ahead with IDs that were never created. Database errors are flattened to their underlying cause so the agent gets an actionable message.

Beyond native tools

The agent's toolbox also includes, when configured:

  • Composio integrations — one tool per connected third-party action (Twitter, LinkedIn, etc.), available once an account is connected.
  • Extension MCP tools — namespaced ext__<extension>__<tool>, connected per request from each enabled extension's MCP server.
  • Extension actions marked exposeAsTool — surfaced as ext_<ext>_<action>.

See Integrations and Extensions.