Skip to main content

AI Agents

AI Agents are configurable automations in FunnelStory that run a series of steps: pull data from your workspace or connections, call an LLM where you need judgment or language, then take action (Slack, email, tasks, CRM updates, datasets, and more). You choose when they run—on a schedule, when events occur, when SQL returns matching rows, or only when someone starts a run manually or from chat.

Build with AI

You can design and refine agents with Cursor, Claude, or any MCP client. See Vibe coding.

How it works​

Each agent is backed by a configuration (a directed graph): an entry step, operations on each node, and variables that pass results between steps. Triggers decide when the platform starts a new run; each run records events so you can inspect what happened.

Core concepts​

ConceptMeaning
TriggerWhen a new run is created (manual, schedule, interval, activity, signal, needle mover, conversation, or query). Published agents use draft: false and a trigger so the runner can enqueue work.
Step graphSteps wired with next; each step has an operation (CALL, AGENT, LOOP, CONDITION, BRANCH, JOIN, TRANSFORM, WAIT, SPAWN).
CALLInvokes a function (for example semantic.query, slack.send_message) with arguments you configure.
AGENTRuns an LLM with optional tools (a subset of the same functions) for multi-step reasoning inside one step.
VariablesStep outputs go to local or global variables; templates interpolate values into strings and SQL.
RunOne execution of the graph from trigger or manual start through completion, failure, or wait states.

Two ways to build​

  1. Canvas — Open Agents, create an agent, set the trigger, add blocks on the canvas, configure each step in the side panel, test, then save and activate.
  2. Vibe coding — Connect an AI assistant to the FunnelStory MCP server; it can read the flow authoring guide, create or update agents, and run them for you.