Automate workflows with AutoGen in Forjinn

AutoGen lets you build automated multi-agent workflows for task completion, code generation, and research — all visually configured from Forjinn's canvas.

AutoGen is a multi-agent framework designed for automated task completion. Agents in an AutoGen flow can converse with each other, write and execute code, browse the web, and coordinate in group chats — all without constant human direction. Forjinn brings AutoGen's capabilities to a visual canvas so you can compose these workflows without writing framework code directly.

Use cases

AutoGen flows work especially well for:

Automated task completion — break a complex goal into sub-tasks distributed across specialized agents
Code generation and review — an assistant agent writes code, a user proxy agent executes it, and the conversation continues until the result is correct
Research pipelines — a web-surfing agent gathers information while an analysis agent synthesizes findings
Structured conversations — round-robin or selector-based group chats where multiple agents take turns contributing

Create an AutoGen flow

In the left sidebar, click AutoGen. The dashboard lists your existing AutoGen flows.

Click Add New and give your flow a name.

Add a Start node to define the initial input and entry point for the flow.

Choose the agent types that match your workflow. Connect agents to each other or into a group chat node.

Select an agent node and fill in its settings — name, system message, model client, tools, and termination behavior.

Add a Termination Condition node and connect it to your group chat or agent. This tells AutoGen when to stop the conversation.

Click Validate to confirm the graph is complete, then Execute to run the flow or Deploy to publish it as an endpoint.

Available node types

Agents

Assistant Agent

A pre-configured assistant that responds to requests. Takes a system message, model client, and optional tools. Human input mode defaults to TERMINATE — the agent asks for human review when it wants to end the conversation.

User Proxy Agent

Acts on behalf of the user. Can execute code, call functions, and interact with the environment. Often paired with an Assistant Agent to form a working code-generation loop.

Conversable Agent

The base agent type with full control over all AutoGen agent settings — useful when neither the assistant nor user proxy presets match your requirements.

Multimodal Agent

An agent that accepts both text and image inputs, suitable for visual reasoning tasks.

Web Surfer Agent

A browser-equipped agent that can search the web, navigate pages, and extract information for use in the conversation.

Group chats

Node	Description
RoundRobin GroupChat	Agents take turns speaking in a fixed rotation. Each agent gets one turn before the cycle repeats.
Selector GroupChat	A selector model picks which agent speaks next based on the current conversation context.
Swarm Team	A swarm-style coordination where agents self-organize around tasks.

Coordination and control

Node	Description
Start	Entry point that defines the initial user message or trigger input.
Termination Condition	Specifies when the conversation should end — by keyword, turn count, or custom logic.
Handoff Config	Configures how an agent hands off control to another agent.
Nested Chat	Embeds one AutoGen conversation inside another as a tool call.

Tools and execution

Node	Description
Function Tool	A custom Python function exposed as a tool for agents to call.
Tools	General-purpose tool selection for attaching capabilities to agents.
Code Execution Config	Configures the sandboxed environment (Docker or local) used to execute agent-generated code.
Model Client	Defines the LLM provider and model settings shared across agents in the flow.

Agent configuration reference

When you configure an Assistant Agent or Conversable Agent, the key settings are:

Setting	What it does
Name	Unique identifier for the agent — used in group chat speaker selection
System Message	The agent's instructions and persona
Model Client	The LLM powering the agent
Tools	Capabilities the agent can invoke
Human Input Mode	`NEVER` — fully automated; `TERMINATE` — asks when ending; `ALWAYS` — asks every turn
Max Consecutive Auto Reply	Caps how many times the agent replies in a row without human input
Is Termination Message	A function that returns true when the conversation should stop
Code Execution Config	Enables code execution with Docker or direct execution settings

AutoGen flows support Python code generation. After building your canvas, click Generate Python to export the equivalent AutoGen Python script. This is useful for running the flow outside Forjinn or understanding exactly what the canvas configuration produces.

For straightforward code-generation tasks, start with an Assistant Agent paired with a User Proxy Agent and a Termination Condition that watches for "TERMINATE" in messages. This two-agent pattern handles most automated programming tasks out of the box.