Product

Agentic Workflows

Multi-step reasoning with tool calling

TL;DR

Multi-step reasoning where agents use tools to complete tasks. Non-agentic (standard RAG):

Key Takeaways

  • Agentic vs Non-Agentic
  • Tool Types
  • Example Use Case
  • Enable Agentic Workflows
  • Tool Configuration
  • When to Use Agentic Workflows

Multi-step reasoning where agents use tools to complete tasks.

Agentic vs Non-Agentic

Non-agentic (standard RAG):

  • Retrieve documents → generate response
  • Single LLM call
  • No external actions

Agentic (with tools):

  • Retrieve documents → decide if tool needed → call tool → synthesize result
  • Multiple LLM calls (planning, execution, synthesis)
  • Can execute actions (API calls, calculations, code)

Requires: Cypress RAG strategy

Tool Types

Built-in tools:

  • search_knowledge_base(query): Recursive retrieval
  • calculate(expression): Math evaluation (Python eval)

Custom tools (Enterprise):

  • api_call(endpoint, method, params): HTTP requests to your APIs
  • run_code(script): Server-side JavaScript execution (sandboxed)

Configure: Agents → [Agent] → Configuration → Tools tab

Example Use Case

Problem: "When is my next payment date?"

Non-agentic response: "Payment dates are on the 1st and 15th of each month."

Agentic response:

  1. Agent retrieves payment schedule from docs
  2. Calls calculate tool with current date
  3. Determines next date: "January 15, 2024"
  4. Responds: "Your next payment date is January 15, 2024."

Observable behavior: Response shows "Tools used: calculate" in debug panel

Enable Agentic Workflows

Location: Agents → [Agent] → Configuration → Advanced Settings

  1. Toggle Agentic Workflow → On
  2. Select tools from list (checkboxes)
  3. Configure tool parameters (if needed)
  4. Click Save

Requirements:

  • RAG strategy must be Cypress
  • Enterprise plan (custom tools)
  • API endpoints must be accessible from Twig servers

Latency impact: +1-3s per tool call (depends on tool execution time)

Tool Configuration

Custom API tool example:

{
  "name": "get_customer_balance",
  "description": "Get customer account balance",
  "endpoint": "https://api.company.com/customers/{customer_id}/balance",
  "method": "GET",
  "auth": {
    "type": "bearer",
    "token": "{{API_TOKEN}}"
  },
  "parameters": {
    "customer_id": {
      "type": "string",
      "description": "Customer ID from conversation context"
    }
  }
}

Observable behavior: Agent detects need, calls API, incorporates result in response

When to Use Agentic Workflows

Use when:

  • Need real-time data (balances, statuses, dates)
  • Require calculations or logic
  • Must execute actions (create ticket, send email)
  • Multi-step reasoning needed

Don't use when:

  • Simple document retrieval sufficient
  • Speed is critical (adds latency)
  • No dynamic data needed
  • Budget constrained (costs more)

Constraints

  • Cypress strategy only (Redwood/Cedar don't support tools)
  • Max 5 tool calls per query
  • Tool timeout: 30 seconds
  • No database writes (read-only APIs recommended)
  • Sandboxed execution (limited Node.js APIs for custom code)

Rate limits: Same as standard queries (100 req/min Execute scope)


Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the ask query parameter:

GET /dev/product/advanced/agentic-workflows.md?ask=<question>

The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.

Last updated January 26, 2026