AgentFlow
FreeAbout Complex LLM Workflows from Simple JSON. 
FreeFree tier
About AgentFlow
Agentflow is an open-source Python tool for creating and executing complex LLM workflows defined in simple JSON files. It bridges the gap between conversational chat interfaces (like ChatGPT) and autonomous agents (like Auto-GPT) by offering step-by-step, reliable workflow execution. Users write workflows in plain English within human-readable JSON, pass dynamic variables for user-specific outputs, and build custom Python functions to go beyond text generation. The tool runs from the command line, requires an OpenAI API key, and is licensed under the MIT License.
Key Features
Define workflows in plain English using human-readable JSON files
Support for dynamic variables to customize outputs per user input
Build and execute custom Python functions for extending beyond text generation
Step-by-step workflow execution guided by LLMs
Command-line interface with verbose mode for real-time task visibility
System message support to guide model behavior
Configurable settings per task, including temperature and function calls
Pros & Cons
Pros
- Fills the gap between conversational chat and fully autonomous agents
- Workflows are predictable and reliable due to step-by-step execution
- Simple JSON configuration makes workflows easy to write, read, and version control
- Extensible with custom Python functions for advanced logic
- Open-source (MIT license) and free to use
- Lightweight and runs locally with minimal dependencies
Cons
- Requires an OpenAI API key, locking users to OpenAI models
- Currently in development with no stable releases or version tags
- Command-line only interface, no graphical user interface
- Limited documentation beyond the README file
- No built-in support for other LLM providers like Anthropic or local models
- Does not include pre-built integrations or a library of workflow templates
Best For
Automating multi-step LLM tasks that require consistency and structureBuilding reliable pipelines for content generation, data processing, or analysisReplacing unreliable autonomous agent workflows with deterministic step-by-step flowsPrototyping and testing complex LLM interactions without building a full application
FAQ
What is Agentflow?
Agentflow is an open-source tool for creating complex LLM workflows using simple JSON files. It executes tasks step-by-step, guided by an LLM.
How do I install Agentflow?
Clone the repository, create a .env file from example.env with your OpenAI API key, and run pip install -r requirements.txt to install dependencies.
Can I use Agentflow with models other than OpenAI?
Currently, Agentflow is built to work with the OpenAI API and requires an API key. There is no built-in support for other providers.
Does Agentflow support custom functions?
Yes. You can create custom Python functions by creating a file in the functions folder, inheriting from BaseFunction, and implementing get_definition() and execute().
Is Agentflow free?
Yes, Agentflow is open-source under the MIT License and free to use. You only need your own OpenAI API key for LLM calls.