Pipelex/pipelex

What it solves

Pipelex addresses the complexity of building, running, and sharing AI workflows. It replaces fragmented, hard-to-maintain scripts with a declarative, typed system for creating "AI Methods"—reusable procedures that ensure structured outputs and repeatable execution across different LLM providers.

How it works

Users define AI procedures in .mthds files using a TOML-like syntax. These files specify inputs, outputs, and the logic (pipes) required to achieve a result. Pipelex handles the underlying orchestration, including:

• Model Routing: Access to 60+ models via a single gateway or local providers (Ollama, vLLM).
• Typed Concepts: Semantic types that ensure data connects correctly between steps.
• Pipeline Orchestration: The ability to chain pipes into sequences, nest methods, and perform batch operations (e.g., processing multiple PDFs).
• Execution Targets: Methods can be run via a CLI, Python SDK, TypeScript SDK, REST API, or as tools for agents via the Model Context Protocol (MCP).

Who it’s for

• AI Developers who want to build portable, structured AI pipelines without being locked into a single model provider.
• Software Engineers looking to integrate AI procedures into existing Python or TypeScript applications.
• Agent Builders who need to provide LLMs with reliable, typed tools.

Highlights

• Declarative Syntax: Human-readable .mthds files make AI logic transparent and easy to edit.
• Multi-Model Support: Seamlessly switch between major providers (OpenAI, Anthropic, Google) and local LLMs.
• Structured Data: Strong typing for inputs and outputs prevents "hallucinated" formats and ensures reliability.
• Extensive Ecosystem: Includes a VS Code extension for visualization, a community hub for sharing methods, and plugins for Claude Code.