Build AI agents that think in code

A framework for building reliable AI agents that express their plans as code and execute it in our custom built AST runtime: enabling stop/start workflows to keep humans in loop, tool-level state management, and enhanced observability.

Star on GitHub

Christopher Lennan

Lead Machine Learning Engineer

"AgentScript brings a refreshing approach to AI development, making it easier to build and deploy intelligent agents."

Powerful Features

Transform your AI development with cutting-edge features designed for advanced agent workflows.

Dynamic Code Generation

LLMs express execution plans as readable JavaScript code

Built-in integration with deterministic and LLM-powered tools

Local variables for flexible data manipulation and passing

AST Transformation

Converts generated JavaScript into Abstract Syntax Tree structure

Runtime interpretation enables controlled and safe execution

Supports key JavaScript features like loops, conditionals, and functions

State Management

Pause and resume execution at any point in the workflow

Serializable state for persistent storage and retrieval

Built-in durability with automatic failure handling

Runtime Environment

Runs in both Node.js and serverless environments

No sandbox needed - secure AST-based interpretation

Native support for tool state and execution tracking

Framework Comparison

See how AgentScript compares to other popular AI agent frameworks

Core Architecture

AgentScript

Code-first with AST transformation
Native code execution
Controlled runtime environment

CrewAI

Role-based with process workflows
Flow-based execution
Team-oriented design

AutoGen

Event-driven actor model
Distributed system design
Multi-language support

LangGraph

State machine with DAG
Graph-based workflows
State transitions

Stay Updated

Join our waitlist to get early access and updates on AgentScript's development.