A lightweight, model-agnostic framework for building AI agent workflows in TypeScript.
AgentForge lets you run agents that can think, use tools, and collaborate on tasks. It’s inspired by systems like Claude Code, but designed to be flexible, minimal, and not tied to a single model provider.
Build agents that can plan, execute, and interact with your environment.
AgentForge is a work-in-progress framework for:
- Running LLM-powered agents
- Giving agents access to tools like the filesystem and shell
- Coordinating multiple agents on shared goals
- Executing tasks with simple dependency handling
It’s meant to be easy to understand, easy to modify, and useful for experimentation.
-
Model agnostic Use different models across agents or plug in your own.
-
Tool-enabled agents Let agents interact with real environments, not just text.
-
Multi-agent workflows Explore agents collaborating instead of working in isolation.
-
Minimal and hackable No heavy abstractions. The goal is clarity and control.
- Automating development workflows
- Prototyping agent-based systems
- Building internal tools powered by LLMs
- Experimenting with multi-agent coordination
This project is actively being developed. Expect rough edges and breaking changes.
MIT