Overview
Build, test, and run multi-step AI workflows visually, no code required. Connect prompts, models, and agents on a drag-and-drop canvas.
About
Agent Playground is Future AGI’s workflow builder for AI agents. It lets you design multi-step AI workflows by dragging nodes onto a canvas and connecting them, no code required.
Most AI applications are not a single prompt. They chain steps together: call one model, pass its output to another, combine results, and so on. As these pipelines grow, keeping track of every connection and debugging failures across steps gets harder. When something breaks, tracing which step went wrong means digging through logs.
Agent Playground makes this visual. You build your workflow as a graph of connected steps. Each step is a node: like an LLM call or a sub-agent. You draw connections between nodes to define how data flows from one step to the next. When you are ready, you hit Run and watch each step execute in real time, with results visible per node.
Key capabilities:
- Visual builder: drag-and-drop canvas to design workflows without writing code
- Real-time execution: run your workflow and watch each step light up as it completes
- Version control: draft changes safely, activate when ready, roll back if needed
- Batch testing: connect a dataset and run your workflow against hundreds of inputs at once
- Reusable components: embed one workflow inside another for modular, composable designs
- Full traceability: every run is recorded with complete input/output details per step
How Agent Playground Connects to Other Features
- Prompt: LLM Prompt nodes use prompts you have already built in the Prompt Management system. Update a prompt once, and every workflow using it picks up the change automatically.
- Dataset: Each graph has a linked dataset where you can set up input variables and run experiments. Go to Dataset to add rows, fill in values for each input, and execute your workflow across all of them.
Know the Parts
Before diving in, here is what each term means and how they fit together.
Graph: your workflow
A graph is the container for your entire workflow. Think of it as a project: it has a name, description, team collaborators, and one or more saved versions. You build and run workflows inside a graph.
Node: a single step
A node is one step in your workflow. There are two kinds:
- LLM Prompt nodes call a language model using a prompt template you have set up in Prompt Management. You pick the model, set parameters like temperature, and the node handles the rest.
- Agent nodes embed an entire other workflow as a single step, useful for breaking complex pipelines into reusable building blocks.
Edge: a connection between steps
An edge is the line connecting two nodes. It defines how data flows from one step to the next. You create edges by dragging from one node’s output to another node’s input on the canvas.
Version: a saved snapshot
Versions let you iterate safely. Make changes in a Draft, then Activate it when you are happy with the result. Previous versions are saved, so you can always go back and pick up from an earlier state.
Execution: a single run
An execution is one run of your workflow. It records the status of every step (success, failed, running), how long each took, and what data went in and came out. You can browse past executions to debug issues.
Getting Started
Core Concepts
Learn how graphs, nodes, and connections work together to form workflows.
Versions & Execution
Understand the version lifecycle and how workflows run.
Create a Graph
Create your first workflow and start building.
Build a Workflow
Add steps, configure them, and connect them into a pipeline.
Run & Monitor
Execute workflows, watch results in real time, and browse history.