llm-exe

llm-exe is a TypeScript package designed to help developers build type-safe AI applications and LLM agents using modular components. It addresses the challenge of constructing structured, reliable AI workflows by allowing users to define typed LLM functions and compose them into reusable executors. The tool is provider-agnostic, supporting integration with various large language model services such as OpenAI, Anthropic, AWS Bedrock, Google, xAI, and Ollama, enabling developers to switch between providers with minimal code changes.

A key feature of llm-exe is its full TypeScript support, which ensures type safety and inferred types throughout LLM chains, reducing uncertainty about data structures during development. The package includes built-in mechanisms for retries, timeouts, error handling, and schema validation, aiming for production readiness and reliability. Developers can chain executors as modular building blocks, each focused on a single task, to create complex workflows. The tool also offers powerful parsers that can extract specific output formats such as JSON, lists, regular expressions, or markdown blocks, with guarantees on output structure or error handling if the format is not met.

llm-exe facilitates the creation of autonomous agents by providing built-in state management, tool calling, and dialogue tracking capabilities. Functions can be turned into agent capabilities, allowing for flexible and extensible agent behaviors. The platform is delivered as a TypeScript package, making it accessible for developers working within the JavaScript and TypeScript ecosystem. It does not require users to manage their own infrastructure for state or chat message storage, as features like DialogueDB are available.

evidence_sufficient=true