Pipelex

Executable AI Methods

Declare multi-step AI methods in typed .mthds files — deterministic orchestration, structured outputs, repeatable results.

Demo (2min) - Docs - Hub - Discord

[pipe.summarize_article]
type    = "PipeLLM"
inputs  = { article = "Text", audience = "Text" }
output  = "Text"
prompt  = "Summarize $article in three bullet points for $audience."

From here, Pipelex handles model routing across 60+ models, structured output parsing, and pipeline orchestration.

A method is written once in the MTHDS open standard and runs anywhere Pipelex does: on your machine with the pipelex runtime, on a server you host with pipelex-api, or on the hosted Pipelex API.

Open Claude Code and paste this. Claude does the rest.

Install mthds: run bash -c 'npm install -g mthds && mthds-agent bootstrap && claude plugin marketplace add mthds-ai/mthds-plugins && claude plugin install mthds@mthds-plugins'. Then tell me to /reload-plugins.

Then build your first method:

/mthds-build A method to summarize articles with key takeaways for different audiences

Run it:

/mthds-run

This installs the local toolchain and runs methods on your machine. To run them on the hosted Pipelex API instead, install pipelex-plugins — nothing to install locally beyond the plugin.

Open Codex and paste this. Codex does the rest.

Install mthds: run bash -c 'npm install -g mthds && mthds-agent bootstrap && mthds-agent codex apply-config && codex plugin marketplace add mthds-ai/mthds-plugins'. Then tell me to restart Codex and run /plugins to install mthds.

Requires Codex 0.130.0+ (npm install -g @openai/codex@latest if needed). See the mthds-plugins README for what mthds-agent codex apply-config does and manual install steps.

uv tool install pipelex
pipelex init

1. Install the VS Code extension for .mthds syntax highlighting — Cursor and other VS Code forks get it from Open VSX
2. Browse methods on the MTHDS Hub and in the public method library for inspiration
3. Author your own .mthds methods
4. Validate with pipelex validate bundle your_method.mthds
5. Run with pipelex run bundle your_method.mthds

A method from the public library runs by its address, with nothing to install first:

pipelex run method github.com/Pipelex/methods/doc_summarizer --inputs inputs.json

The hosted Pipelex API at api.pipelex.com runs methods durably — start a run, poll it, fetch the result — with file storage and a method catalog on top. It is currently in private beta: join the waitlist, then get an API key at app.pipelex.com. The SDKs also speak to a self-hosted pipelex-api for its protocol routes, such as execute, validate, and codegen, when PIPELEX_BASE_URL points at your instance; durable runs, file storage, and the method catalog are the hosted API's own, and the starter templates and plugins are built on them.

Three ways in:

From a coding agent — pipelex-plugins. Skills and hooks for Claude Code, Codex, and Mistral Vibe, with no local toolchain: methods are validated and run through the Pipelex MCP server, which the plugin launches for you.

claude plugin marketplace add Pipelex/pipelex-plugins
claude plugin install pipelex@pipelex-plugins

For Codex, codex plugin marketplace add Pipelex/pipelex-plugins, restart, then /plugins to install pipelex (Codex 0.141+). Then /pipelex-design A method to … builds a method, /pipelex-inputs prepares its inputs and offers to run it, and /pipelex-integrate wires it into your codebase with generated types.

From any MCP host — @pipelex/mcp. One capability core, two servers: a local workshop that coding agents spawn with npx and that reads .mthds files straight from disk, and a hosted console you add as a custom connector in ChatGPT, claude.ai, or Claude Desktop, with nothing to install.

claude mcp add pipelex --env PIPELEX_API_KEY=plx_sk_... -- npx -y @pipelex/mcp

From your own code — the SDKs. @pipelex/sdk for TypeScript and pipelex-sdk for Python cover the MTHDS Protocol routes, the durable run lifecycle, storage, and the product surface, and both ship an offline check that keeps generated types in sync with your methods.

npm install @pipelex/sdk
pip install pipelex-sdk

Two ready-to-fork templates, one per language, both calling the Pipelex API. Click Use this template on GitHub, then run the bundled /bootstrap skill in Claude Code to make the project yours.

• pipelex-starter-js — Next.js 16 + TypeScript app calling the Pipelex API via @pipelex/sdk. Best when you want a TypeScript frontend or backend that talks to the hosted Pipelex API. Ships with demo pipelines, from text entity extraction and PDF summaries to image generation, plus a method pulled in from the public library by address.
• pipelex-starter-python — Python CLI calling the Pipelex API via pipelex-sdk, with no local runtime to install. Best when you want a script, service, or command-line tool that runs methods remotely and prints structured JSON with a cost report.

To run methods in-process from Python instead, install the pipelex runtime and start from the Cookbook.

• Pipelex Gateway (Recommended) — Free credits, single API key for LLMs, OCR / document extraction, and image generation across all major providers.
• Bring Your Own Keys — Use existing API keys from OpenAI, Anthropic, Google, Mistral, etc. See Configure AI Providers.
• Local AI — Ollama, vLLM, LM Studio, or llama.cpp — no API keys required. See Configure AI Providers.

• Cookbook — Ready-to-run methods for the pipelex runtime: classification, extraction, analysis, generation, and more.
• Public method library — Packaged methods you run by address, from document extraction and invoice processing to slide design and image generation.

• Discord — Get help, share methods, meet the team
• Documentation — Guides and reference
• GitHub Issues — Report bugs and request features
• security@pipelex.com — Security and privacy concerns

Runtime and servers

Agent plugins

SDKs and starters

Methods and examples

The MTHDS standard

Tooling and libraries

The runtime and the servers that run methods — pipelex, pipelex-api and pipelex-mcp — are source-available under the Elastic License 2.0 (ELv2). You may embed them in your own products, including services whose features run your methods, and run them for your own team or company; what the license rules out is hosting a service that runs methods for others. The license page explains how we read it, with concrete examples. Every version of them released before the switch stays under MIT.

The SDKs, the starter templates, the public method library, the cookbook, the UI libraries, the VS Code extension, the n8n node, cocode and the MTHDS open standard are MIT licensed. kajson and pipelex-plugins are licensed under Apache 2.0. Each repository's LICENSE file carries its terms.

"Pipelex" is a trademark of Evotis S.A.S.