ai-agents Archives - Vijay Gokarn https://vijay-gokarn.com/tag/ai-agents/ "Ignite Curiosity. Fuel the Future." Sat, 18 Apr 2026 11:20:04 +0000 en-US hourly 1 https://wordpress.org/?v=6.9.4 https://i0.wp.com/vijay-gokarn.com/wp-content/uploads/2023/09/cropped-ideogram.jpeg?fit=32%2C32&ssl=1 ai-agents Archives - Vijay Gokarn https://vijay-gokarn.com/tag/ai-agents/ 32 32 230943525 Creating AI Storytelling Agents Using Flowise: A Step-by-Step Guide https://vijay-gokarn.com/creating-ai-storytelling-agents-using-flowise-a-step-by-step-guide/?utm_source=rss&utm_medium=rss&utm_campaign=creating-ai-storytelling-agents-using-flowise-a-step-by-step-guide Fri, 24 Jan 2025 16:51:01 +0000 https://vijay-gokarn.com/?p=227 GenAI Mastery Series · Agentic AI · Flowise Walkthrough Building an AI Storytelling Agent with Flowise — No Code Required StackFlowise · OpenAI GPT-4 · Supervisor/Worker Nodes DeploymentLocal · Cloud-ready OutputIBM the Robot’s Marshmallow Party Concepts Covered AI Agents Flowise Workflows Supervisor / Worker Pattern ChatOpenAI Node No-Code Orchestration Prompt Engineering In today’s AI landscape, […]

The post Creating AI Storytelling Agents Using Flowise: A Step-by-Step Guide appeared first on Vijay Gokarn.

]]>

GenAI Mastery Series · Agentic AI · Flowise Walkthrough

Building an AI Storytelling Agent with Flowise — No Code Required

StackFlowise · OpenAI GPT-4 · Supervisor/Worker Nodes

DeploymentLocal · Cloud-ready

OutputIBM the Robot’s Marshmallow Party

Concepts Covered AI Agents Flowise Workflows Supervisor / Worker Pattern ChatOpenAI Node No-Code Orchestration Prompt Engineering

In today’s AI landscape, agents are becoming powerful tools to automate complex tasks — from chatbots to interactive storytelling. Flowise is a no-code AI workflow builder that makes it easy to design, deploy, and manage AI agents for a wide range of applications. This walkthrough builds a fully functional storytelling agent, locally deployable and cloud-ready.

Concepts

What are AI agents in Flowise?

AI agents in Flowise are intelligent modules that can handle tasks autonomously by combining logic, AI models, and external tools. They process inputs, make decisions, and generate tailored outputs — without manual intervention at each step.

In this project we use the Supervisor and Worker node pattern with OpenAI Chat. The supervisor coordinates the overall workflow; worker nodes each own a specific sub-task — here, storytelling and title assignment.

Why multi-agent? Splitting responsibilities between nodes keeps each prompt focused and small. A dedicated Storyteller node generates better stories than one giant prompt trying to write a story, title it, and format it all at once. This mirrors how real engineering teams work — one job per role.

Step 1

Setting up Flowise

Flowise runs as a local Node.js server you access through a browser-based canvas. Two commands are all you need to get started.

Terminal — Install & Run npm install flowise npx flowise run

Once running, open your browser, log in, and click “New Workflow” to open the interactive canvas. You’ll drag, drop, and wire nodes together visually — no boilerplate code.

Local vs Cloud: The setup above runs entirely on your machine. For cloud deployment, Flowise supports Railway, Render, and self-hosted Docker. The workflow JSON is portable — build locally, deploy anywhere.

Step 2 — Workflow Design

Building the agent — node by node

1

Chat Model Node

ChatOpenAI — The Brain

Drag a ChatOpenAI Node onto the canvas and connect it as the model backend for all worker nodes. Configure GPT-4 with elevated temperature for imaginative outputs.

Model: GPT-4 Temperature: 0.9 Max Tokens: 400–500
2

Worker Node · Role: Storyteller

Storytelling Agent

Add a Worker Node and connect it to the ChatOpenAI node. Set its role as the Storyteller. This node owns the core creative generation task — it receives the theme prompt and writes the full story.

Worker Prompt You are a storyteller. Write a fun and engaging story for kids aged 5–8. The main character is a robot named IBM. Make it funny, magical, and include a twist. Limit the story to 400 words.
3

Worker Node · Role: Title Assigner

Title Assigner Agent

Add a second Worker Node downstream of the Storyteller. This node’s sole job is to extract a short, engaging title from the generated story — a focused single-responsibility task.

Worker Prompt Extract the title of the story you just created. Keep it short and engaging.
4

Format Prompt Values Node

Output Formatter

Use the Format Prompt Values Node to combine the story and title from the two worker nodes into a clean, structured output ready for display.

Title: {‘{‘}Title Extracted{‘}’} Story: {‘{‘}Generated Story{‘}’}
5

Chat Output Node

Chat Output — Delivery

Connect the formatted output to the Chat Output Node. This is the interface layer — the final assembled story and title are surfaced here for users to read, copy, or embed.

Displays story + title Embeddable chat widget

Step 3

Running the agent

With all nodes configured and wired together, save your workflow — name it something like “AI Storytelling Agent” — and hit Run. Enter a theme prompt or use the default storytelling instructions, and the agent pipeline fires automatically: ChatOpenAI powers the Storyteller worker, its output flows to the Title Assigner, both outputs merge in the Formatter, and the Chat Output displays the result.

Tip: Experiment with different temperature values. At 0.7 the stories are coherent but predictable. At 0.95 you get genuinely surprising plot twists — which for kids’ stories is exactly what you want.

Example Output

IBM’s Marshmallow Party

Generated Title

IBM the Robot and the Marshmallow Party

Once upon a time, in a land of giggles and sparkles, there lived a silly little robot named IBM. Now, IBM wasn’t your everyday robot who danced or painted; oh no! He was known as a “computer,” which is a magic box that helps people do all sorts of amazing things!

One sunny day, IBM decided to throw a party for all his robot friends. He said, “Let’s make it the best party ever! I’ll invite my buddy, Printer Pete, and my bestie, Codey the Coder!” But IBM accidentally ordered 1,000 bags of rainbow-colored marshmallows instead of snacks — and chaos ensued!

The party turned into a marshmallow-filled adventure with pillow fights, marshmallow towers, and lots of laughs. In the end, IBM and his friends agreed: whether it’s chips or marshmallows, any party is fun when friends are around!

ModelGPT-4

Temperature0.9

Word Count~150 words

Nodes Used5

Interview Prep

Cheat sheet — quick definitions to remember

Define
What is an AI agent?
An autonomous module that combines a language model, logic, and optionally external tools to complete a task without step-by-step human instruction. It receives a goal, plans sub-steps, and executes them independently.
AutonomousGoal-drivenTool-using
Explain
What is the Supervisor / Worker pattern?
A Supervisor node coordinates the overall workflow and delegates tasks to Worker nodes, each of which handles one focused sub-task. This mirrors microservices architecture — single responsibility per agent, composable into larger pipelines.
Supervisor = orchestratorWorker = specialistSingle responsibility
Define
What does temperature control in an LLM?
Randomness in token sampling. Low temperature (0.1–0.4) = deterministic, factual, conservative outputs. High temperature (0.8–1.0) = creative, surprising, occasionally incoherent. For storytelling, 0.9 hits the sweet spot of imaginative without losing coherence.
Low = deterministicHigh = creative0.9 for stories
Compare
No-code (Flowise) vs code-first (LangChain) — when to use which?
Use Flowise for rapid prototyping, demos, non-developer stakeholders, or when the workflow is straightforward and visual. Use LangChain / LangGraph in code when you need version control, CI/CD, complex branching, custom tool integrations, or production-grade observability.
Flowise = prototype fastLangChain = production
Explain
Why split story generation and title extraction into separate nodes?
Focused prompts outperform omnibus prompts. A prompt that must write a story, extract a title, and format output all at once tends to trade off quality across tasks. Separate nodes give each sub-task its own context window, model parameters, and success criteria — and makes each step independently testable and replaceable.
One node, one jobBetter qualityIndependently testable
Use Case
What other use cases suit a Flowise multi-agent setup?
Any pipeline with distinct sequential sub-tasks: customer support (intent classification → knowledge retrieval → response drafting), content pipelines (research → outline → write → SEO optimize), data workflows (extract → validate → transform → summarize).
Customer supportContent pipelinesData workflowsCode review agents
Gotcha
What are the main limits of no-code agent builders?
Three key limitations: (1) Observability — debugging visual workflows is harder than reading stack traces. (2) Version control — workflow JSON doesn’t diff cleanly in Git. (3) Custom logic — complex conditional branching, stateful memory, and custom tool integrations are much easier in code-first frameworks.
Hard to debugNo clean Git diffLimited branching

GenAI Mastery Series — vijay-gokarn.com · Vijay Gokarn

Back to Blog ↗

GenAI Mastery Series · Agentic AI · Flowise Walkthrough

Building an AI Storytelling Agent with Flowise — No Code Required

StackFlowise · OpenAI GPT-4 · Supervisor/Worker Nodes

DeploymentLocal · Cloud-ready

OutputIBM the Robot’s Marshmallow Party

Concepts Covered AI Agents Flowise Workflows Supervisor / Worker Pattern ChatOpenAI Node No-Code Orchestration Prompt Engineering

In today’s AI landscape, agents are becoming powerful tools to automate complex tasks — from chatbots to interactive storytelling. Flowise is a no-code AI workflow builder that makes it easy to design, deploy, and manage AI agents for a wide range of applications. This walkthrough builds a fully functional storytelling agent, locally deployable and cloud-ready.

Concepts

What are AI agents in Flowise?

AI agents in Flowise are intelligent modules that can handle tasks autonomously by combining logic, AI models, and external tools. They process inputs, make decisions, and generate tailored outputs — without manual intervention at each step.

In this project we use the Supervisor and Worker node pattern with OpenAI Chat. The supervisor coordinates the overall workflow; worker nodes each own a specific sub-task — here, storytelling and title assignment.

Why multi-agent? Splitting responsibilities between nodes keeps each prompt focused and small. A dedicated Storyteller node generates better stories than one giant prompt trying to write a story, title it, and format it all at once. This mirrors how real engineering teams work — one job per role.

Step 1

Setting up Flowise

Flowise runs as a local Node.js server you access through a browser-based canvas. Two commands are all you need to get started.

Terminal — Install & Run npm install flowise npx flowise run

Once running, open your browser, log in, and click “New Workflow” to open the interactive canvas. You’ll drag, drop, and wire nodes together visually — no boilerplate code.

Local vs Cloud: The setup above runs entirely on your machine. For cloud deployment, Flowise supports Railway, Render, and self-hosted Docker. The workflow JSON is portable — build locally, deploy anywhere.

Step 2 — Workflow Design

Building the agent — node by node

1

Chat Model Node

ChatOpenAI — The Brain

Drag a ChatOpenAI Node onto the canvas and connect it as the model backend for all worker nodes. Configure GPT-4 with elevated temperature for imaginative outputs.

Model: GPT-4 Temperature: 0.9 Max Tokens: 400–500
2

Worker Node · Role: Storyteller

Storytelling Agent

Add a Worker Node and connect it to the ChatOpenAI node. Set its role as the Storyteller. This node owns the core creative generation task — it receives the theme prompt and writes the full story.

Worker Prompt You are a storyteller. Write a fun and engaging story for kids aged 5–8. The main character is a robot named IBM. Make it funny, magical, and include a twist. Limit the story to 400 words.
3

Worker Node · Role: Title Assigner

Title Assigner Agent

Add a second Worker Node downstream of the Storyteller. This node’s sole job is to extract a short, engaging title from the generated story — a focused single-responsibility task.

Worker Prompt Extract the title of the story you just created. Keep it short and engaging.
4

Format Prompt Values Node

Output Formatter

Use the Format Prompt Values Node to combine the story and title from the two worker nodes into a clean, structured output ready for display.

Title: {‘{‘}Title Extracted{‘}’} Story: {‘{‘}Generated Story{‘}’}
5

Chat Output Node

Chat Output — Delivery

Connect the formatted output to the Chat Output Node. This is the interface layer — the final assembled story and title are surfaced here for users to read, copy, or embed.

Displays story + title Embeddable chat widget

Step 3

Running the agent

With all nodes configured and wired together, save your workflow — name it something like “AI Storytelling Agent” — and hit Run. Enter a theme prompt or use the default storytelling instructions, and the agent pipeline fires automatically: ChatOpenAI powers the Storyteller worker, its output flows to the Title Assigner, both outputs merge in the Formatter, and the Chat Output displays the result.

Tip: Experiment with different temperature values. At 0.7 the stories are coherent but predictable. At 0.95 you get genuinely surprising plot twists — which for kids’ stories is exactly what you want.

Example Output

IBM’s Marshmallow Party

Generated Title

IBM the Robot and the Marshmallow Party

Once upon a time, in a land of giggles and sparkles, there lived a silly little robot named IBM. Now, IBM wasn’t your everyday robot who danced or painted; oh no! He was known as a “computer,” which is a magic box that helps people do all sorts of amazing things!

One sunny day, IBM decided to throw a party for all his robot friends. He said, “Let’s make it the best party ever! I’ll invite my buddy, Printer Pete, and my bestie, Codey the Coder!” But IBM accidentally ordered 1,000 bags of rainbow-colored marshmallows instead of snacks — and chaos ensued!

The party turned into a marshmallow-filled adventure with pillow fights, marshmallow towers, and lots of laughs. In the end, IBM and his friends agreed: whether it’s chips or marshmallows, any party is fun when friends are around!

ModelGPT-4

Temperature0.9

Word Count~150 words

Nodes Used5

Interview Prep

Cheat sheet — quick definitions to remember

Define
What is an AI agent?
An autonomous module that combines a language model, logic, and optionally external tools to complete a task without step-by-step human instruction. It receives a goal, plans sub-steps, and executes them independently.
AutonomousGoal-drivenTool-using
Explain
What is the Supervisor / Worker pattern?
A Supervisor node coordinates the overall workflow and delegates tasks to Worker nodes, each of which handles one focused sub-task. This mirrors microservices architecture — single responsibility per agent, composable into larger pipelines.
Supervisor = orchestratorWorker = specialistSingle responsibility
Define
What does temperature control in an LLM?
Randomness in token sampling. Low temperature (0.1–0.4) = deterministic, factual, conservative outputs. High temperature (0.8–1.0) = creative, surprising, occasionally incoherent. For storytelling, 0.9 hits the sweet spot of imaginative without losing coherence.
Low = deterministicHigh = creative0.9 for stories
Compare
No-code (Flowise) vs code-first (LangChain) — when to use which?
Use Flowise for rapid prototyping, demos, non-developer stakeholders, or when the workflow is straightforward and visual. Use LangChain / LangGraph in code when you need version control, CI/CD, complex branching, custom tool integrations, or production-grade observability.
Flowise = prototype fastLangChain = production
Explain
Why split story generation and title extraction into separate nodes?
Focused prompts outperform omnibus prompts. A prompt that must write a story, extract a title, and format output all at once tends to trade off quality across tasks. Separate nodes give each sub-task its own context window, model parameters, and success criteria — and makes each step independently testable and replaceable.
One node, one jobBetter qualityIndependently testable
Use Case
What other use cases suit a Flowise multi-agent setup?
Any pipeline with distinct sequential sub-tasks: customer support (intent classification → knowledge retrieval → response drafting), content pipelines (research → outline → write → SEO optimize), data workflows (extract → validate → transform → summarize).
Customer supportContent pipelinesData workflowsCode review agents
Gotcha
What are the main limits of no-code agent builders?
Three key limitations: (1) Observability — debugging visual workflows is harder than reading stack traces. (2) Version control — workflow JSON doesn’t diff cleanly in Git. (3) Custom logic — complex conditional branching, stateful memory, and custom tool integrations are much easier in code-first frameworks.
Hard to debugNo clean Git diffLimited branching

GenAI Mastery Series — vijay-gokarn.com · Vijay Gokarn

Back to Blog ↗

The post Creating AI Storytelling Agents Using Flowise: A Step-by-Step Guide appeared first on Vijay Gokarn.

]]> 227