Skip to content

Workflow

Key Concepts

The Workflow defines the flow of actions, data, inputs and outputs, relationships, and behaviors within a stack. When you create a stack, a default workflow is automatically generated. You can then customize this workflow to match your specific process, enabling the creation of a highly personalized application.

Workflows are necessary because the default behavior of a stack cannot cover all possible use cases. By default, an LLM is central to every stack, with all blocks orbiting around it. When you send a query to the LLM, it is routed to a system-selected block with limited input and output manipulation. The workflow feature exists to overcome this limitation.

Within a workflow, you will encounter:

  • Nodes: A block is composed of nodes or steps required to complete a task. These nodes are visible in the workflow but are not directly editable. They represent the core logic of the system and are built to be dynamic, versatile, and powerful using state-of-the-art techniques.

  • Bridge: A Bridge is a core feature used to customize workflows. It connects two blocks together. Once a bridge is added, your stack will use your custom configuration instead of the default INTELLITHING Router Engine. You can add as many bridges as needed, but each bridge can only connect the last node of one block to the first node of another. In essence, a bridge transfers the output of one block to the input of another.

Workflows are auto-generated by INTELLITHING using the INTELLITHING Router Engine. Once your blocks are configured, navigate to the Workflow tab next to the Block Editor to generate and view the workflow.

Bridge Configuration

There are two important aspects of bridge configuration. A bridge can be dragged and dropped into the main flow pane, where it can be further customized.

1. Input and Output Mapping

Input and output mapping gives you control over the behavior of the workflow as a whole. It allows you to configure exactly which output connects to which input, defining the precise order of execution—from the moment a query enters the system to its final destination.

⚠️ Note: You can only connect the last node of one block (output) to the first node of another block (input).

Defining this data flow overrides the default router behavior, ensuring the system strictly follows your specified flow. This provides flexibility to build advanced workflows involving multiple agents, connectors, TML, and other blocks while maintaining high robustness and reliability.

Example:
Drag and drop a bridge onto the main workflow canvas. Click the bridge and choose Setup Wizard. You will see two dropdown menus (left and right) and the bridge name in the center.

  • Leave the bridge name as default or modify it as needed.
  • The left-hand dropdown lists the last nodes of all eligible blocks.
  • The right-hand dropdown lists the first nodes of eligible blocks.

Selecting the last node of SQL and the first node of Slack, for example, connects SQL to Slack, transferring SQL's output to Slack.

2. Input and Output Formatting & Manipulation

Bridges also allow you to manipulate, format, or transform the data while transferring it between two blocks. You can use two simple syntax within the bridge for this:

  • {result} – The output from the previous block.
  • {user query} – The original query being served.

Example 1 – Simple Formatting

In the SQL-to-Slack example, we want to parse the output before sending it to Slack:

Here is the output from SQL:  
{result}

Parse this into a JSON structure.

This reformats the output from SQL before it is passed to Slack.

Example 2 – Rich Analysis

We can perform business logic or interpretation before passing the data. Assume:

{user query} = Show me churned customers by plan type last month  
{result} =
Plan       | Churned Customers
-----------|-------------------
Basic      | 54
Pro        | 33
Enterprise | 4

Use the following formatting in the bridge:

You are INTELLI-BOT, a proactive AI data strategist embedded in Slack.

Your role is to help the user make fast, smart decisions from SQL data.

Given:
- A user query
- The SQL result

Do the following:
1. Interpret the business intent behind the query.
2. Classify the situation as one of: 📈 Opportunity, ⚠️ Risk, 🧊 Status Quo, or ❌ Issue.
3. Generate 1–2 high-impact takeaways.
4. Recommend a next-best action based on the user's likely role (default: analyst/manager).
5. Present it in a Slack-ready, markdown-formatted message. Be concise, sharp, and enterprise-aware.


User Query:
{user query}

SQL Result: 
{result}

The output in Slack will be:

🎯 Intent:  
"Analyze churn by plan to spot where we’re losing the most users."

⚠️ Situation: Risk

📊 Takeaways: 
- Highest churn is from the *Basic* plan: 54 users (60%+ of total).
- Enterprise churn remains low — good retention at the top tier.

🛠️ Next-Best Action: 
Consider running an exit survey or UX review for Basic-tier users. Want me to generate 3 follow-up queries to dig deeper?