Workflow DSL

Chain Nodes workflows are defined using a declarative YAML schema. This allows for full version control, portability, and human readability.

Basic Structure

apiVersion: chainnodes/v1
kind: Workflow
metadata:
  name: financial-analysis
  version: 1
  description: "Extracts metrics from a financial report and performs validation"

inputs:
  - name: ticker
    type: string
  - name: report
    type: file

blackboard:
  schema:
    ticker: { type: string, required: true }
    revenue: { type: number }
    report: { type: file }
    approved: { type: bool }

states:
  - name: START
    type: initial
    agent: analyst
    instructions: "Read the report and extract the revenue for {{ ticker }}. Store in revenue."

  - name: REVIEW
    type: hitl
    assignee: finance-team
    instructions: "Verify the extracted revenue total."

  - name: TERMINAL
    type: terminal

transitions:
  - from: START
    to: REVIEW
    trigger: success

  - from: REVIEW
    to: TERMINAL
    trigger: approved

Node Types

Type	Description
prompt	Executes an LLM agent with optional tools (MCP).
hitl	Pauses execution for human input or judgment.
script	Executes a small JS expression for data manipulation.
wait	Pauses for a specific duration or external signal.
terminal	The end state of a workflow.

Expressions

Chain Nodes uses the expr engine. You can access the blackboard data using {{ blackboard.key }} in prompts or blackboard.key directly in scripts.

Generative Workflows & Showcase Templates

Chain Nodes supports AI-driven generation that creates fully fleshed out, runnable workflows directly from plain-language Process Descriptions under the Generate tab in the Designer sidebar.

To help showcase and dogfood Chain Nodes's agentic orchestration, we maintain a set of curated showcase templates. These files are stored directly inside the repository under docs/workflows/ and serve as perfect prompt blueprints:

1. GitHub Triage & Auto-Responder: Automatically fetches unread GitHub repository issues, runs root-cause analysis, classifies severity, drafts replies, routes critical issues to a Human-in-the-Loop review pane, and posts approved comments using the GitHub MCP server.
2. Gmail Inbox Sorter & Auto-Drafting: Automatically accesses unread inbox emails via Gmail MCP, assigns an importance rating, drafts professional replies, and prompts the user for review before transmitting.
3. Crypto Market Sentiment Tracker: Fetches statistics for the top 10 traded cryptocurrencies, pulls recent web/social sentiment articles, calculates overall market indices, and synthesizes a comprehensive Markdown report.
4. Executive Brand Watcher & Competitor Radar: Monitors online mentions of brands or competitors, triages sentiment, alerts maintainers of critical negative events, and posts final executive briefs to Slack.
5. Autonomous Invoice & Expense Auditor: Uses LLM vision to extract invoice totals from local folders, performs mathematical validation via a JS code node, and raises a HITL gate on mathematical mismatches.
6. Intelligent Research & Dossier Compiler: Decomposes complex topics, performs parallel Brave and Wikipedia searches, and synthesizes a clean academic Markdown file directly onto the local filesystem.

To try these out, simply copy-paste any template description directly into Chain Nodes Designer's Generate box!