Agentscript

AgentScript is an open-source framework for building AI agents that express their plans as JavaScript code and execute them in a custom AST runtime. Instead of orchestrating agents through chained LLM calls, AgentScript prompts an LLM to generate a script, parses it into an Abstract Syntax Tree (AST), and runs it in a controlled environment. This code-first approach enables stop/start workflows, tool-level state management, and enhanced observability — making it ideal for complex, multi-step tasks that require durability and human-in-the-loop approval. Targeted at developers comfortable with JavaScript, AgentScript supports both Node.js and serverless environments. It integrates with Anthropic's Claude, OpenAI's GPT models, and other LLM tools. Key features include dynamic code generation, AST transformation for safe execution without a sandbox, serializable state for pause/resume, and built-in durability with automatic failure handling. The runtime supports loops, conditionals, functions, and local variables for flexible data manipulation. AgentScript's AST interpretation provides security without needing a sandbox, and its state management allows agents to be paused on any statement or tool call, serialized to a database, and resumed later. This makes the framework robust to interruptions and long-running processes. Developers also benefit from testing utilities and observability with state snapshots. Compared to frameworks like CrewAI or AutoGen, AgentScript's code-centric model offers deeper control and safety for complex, stateful workflows. It is less suited for simple linear pipelines or non-developer users. The project is actively maintained on GitHub with a growing community, and it remains free and open-source under permissive licensing.