Skip to content
Workflows

Orchestrate multi-step AI automations

Workflows chain your agents, tools, and branching logic into repeatable automations. Define the steps once, run them on demand or on a schedule — no glue code, no orchestration infrastructure.

The concept

What is a workflow?

A workflow is a sequence of steps — each one can be an agent turn, a tool call, or a data transformation — connected by inputs and outputs. Steps run in order, and branch points route execution based on what the previous step produced.

Sequential steps

Steps execute one after another. Each step receives the outputs of previous steps as inputs, so context and data flow naturally through the entire workflow without manual wiring.

  • Outputs automatically piped to the next step
  • Typed input/output contracts per step
  • Pause-and-resume on long-running operations

Branching on results

Add a branch node to route execution down different paths depending on what a step produced. Use model output, HTTP status codes, or any structured value as the routing condition.

  • Route on model output, status codes, or values
  • Fallback else branch for unmatched conditions
  • Nested branches supported for complex logic

Chain agents & tools

Mix agent steps and tool steps freely within a single workflow. An agent can call its own tools, and its output feeds directly into the next step — creating powerful multi-model pipelines.

  • Agents and HTTP tools in the same flow
  • Agent output maps to tool input parameters
  • Reuse the same agent across multiple workflows

Step by step

How a workflow runs

From trigger to final output, here is the full lifecycle of a single workflow execution.

  1. 1

    Trigger

    A schedule, webhook event, or manual run kicks off the workflow execution.

  2. 2

    Step 1 — agent

    An AI agent processes the input using its bound model, system prompt, and allowed tools.

  3. 3

    Step 2 — tool

    A tool step (e.g. an HTTP request) fetches, transforms, or writes external data.

  4. 4

    Branch

    A branch node inspects step outputs and routes execution to the matching path.

  5. 5

    Output

    Results are returned to the caller, delivered via webhook, or stored for later review.

Visual workflow editor showing trigger, agent steps, branch node, and output.

Use cases

What can you automate?

Any multi-step task that today requires custom glue code is a strong candidate for a workflow.

Onboarding automation

Trigger a workflow when a new user signs up. An agent personalises the welcome message, a tool call provisions their initial workspace data, and a final step delivers a tailored getting-started guide.

  • Webhook trigger fires on sign-up event
  • Personalised copy generated per user
  • Provisioning tool call runs in the same flow

Report generation

Schedule a workflow to pull metrics from your analytics API, pass them to a report-writer agent, and post the formatted digest to Slack or email — every Monday morning, hands-free.

  • Cron-triggered on any schedule
  • Structured data converted to AI narrative
  • Delivered via HTTP tool to Slack, email, or webhook

Multi-stage data enrichment

Fetch a list of records from your CRM, run each through an enrichment agent that calls external APIs for firmographic data, then write the results back — all in one workflow run.

  • Iterates over lists of input records
  • Per-item agent call with full tool access
  • Write-back step posts enriched data to source

Scheduled digests

Build a daily or weekly digest that scans news feeds, summarises the most relevant articles for your team, and delivers a concise briefing — filtered by the topics you define in the agent prompt.

  • Configurable topic filters in agent system prompt
  • Deduplication across consecutive digest runs
  • Customisable delivery format per recipient

Visual builder

Build a workflow without code

Drag steps from the palette — agent, tool, branch, output — onto the canvas, then connect them. Each step's settings open in a side panel. The canvas shows the full execution graph as you build.

The workflow canvas: trigger node, agent step, branch node, and output step connected by data-flow edges.

Workflows reuse your agents and tools

Every agent step references an agent you have already built — including its bound model, system prompt, and allowed tools. Configure once, reuse everywhere. The same agent can power ad-hoc chats and scheduled workflows simultaneously, with no duplication.

Start building with The Chat Agent

Connect your favourite AI providers, add tools and agents, and ship in minutes. Free to start — no credit card required.