Pipelex is an open-source devtool that transforms how you build repeatable AI workflows. Think of it as Docker or SQL for AI operations.
Create modular "pipes", each using a different LLM and guaranteeing structured outputs. Connect them like LEGO blocks sequentially, in parallel, or conditionally, to build complex knowledge transformations from simple, reusable components.
Stop reinventing AI workflows from scratch. With Pipelex, your proven methods become shareable, versioned artifacts that work across different LLMs. What took weeks to perfect can now be forked, adapted, and scaled instantly.
Pipelex makes it easy for developers to define and run repeatable AI workflows. At its core is a clear, declarative pipeline language specifically crafted for knowledge-processing tasks.
Build pipelines from modular pipes that snap together. Each pipe can use a different language model (LLM) or software to process knowledge. Pipes consistently deliver structured, predictable outputs at each stage.
Pipelex uses TOML syntax, making workflows readable and shareable. Business professionals, developers, and AI coding agents can all understand and modify the same pipeline definitions.
Example:
[concept]
Buyer = "The person who made the purchase"
PurchaseDocumentText = "Transcript of a receipt, invoice, or order confirmation"
[pipe.extract_buyer]
PipeLLM = "Extract buyer from purchase document"
inputs = { purchase_document_text = "PurchaseDocumentText" }
output = "Buyer"
llm = "llm_to_extract_info"
prompt_template = """
Extract the first and last name of the buyer from this purchase document:
@purchase_document_text
"""Pipes are modular building blocks that connect sequentially, run in parallel, or call sub-pipes. Like function calls in traditional programming, but with a clear contract: knowledge-in, knowledge-out. This modularity makes pipelines perfect for sharing: fork someone's invoice processor, adapt it for receipts, share it back.
Pipelex is an open-source Python library with a hosted API launching soon. It integrates seamlessly into existing systems and automation frameworks. Plus, it works as an MCP server so AI agents can use pipelines as tools.
📚 Note that you can check out the Pipelex Documentation for more information and clone the Pipelex Cookbook repository for ready-to-run examples.
Follow these steps to get started:
Visit the
# Using pip
pip install pipelex
# Using Poetry
poetry add pipelex
# Using uv (Recommended)
uv pip install pipelexWe highly recommend installing an extension for TOML files into your IDE of choice. For VS Code, the Even Better TOML extension does a great job of syntax coloring and checking.
The package supports the following additional features:
anthropic: Anthropic/Claude supportgoogle: Google models (Vertex) supportmistralai: Mistral AI supportbedrock: AWS Bedrock supportfal: Image generation with Black Forest Labs "FAL" service
Install all extras:
Using pip:
pip install "pipelex[anthropic,google,mistralai,bedrock,fal]"Using poetry:
poetry add "pipelex[anthropic,google,mistralai,bedrock,fal]"Using uv:
uv pip install "pipelex[anthropic,google,mistralai,bedrock,fal]"domain = "tech_tweet"
definition = "A pipeline for optimizing tech tweets using Twitter/X best practices"
[concept]
DraftTweet = "A draft version of a tech tweet that needs optimization"
OptimizedTweet = "A tweet optimized for Twitter/X engagement following best practices"
TweetAnalysis = "Analysis of the tweet's structure and potential improvements"
WritingStyle = "A style of writing"
[pipe]
[pipe.analyze_tweet]
PipeLLM = "Analyze the draft tweet and identify areas for improvement"
inputs = { draft_tweet = "DraftTweet" }
output = "TweetAnalysis"
llm = "llm_for_writing_analysis"
system_prompt = """
You are an expert in social media optimization, particularly for tech content on Twitter/X.
Your role is to analyze tech tweets and check if they display typical startup communication pitfalls.
"""
prompt_template = """
Evaluate the tweet for these key issues:
**Fluffiness** - Overuse of buzzwords without concrete meaning (e.g., "synergizing disruptive paradigms")
**Cringiness** - Content that induces secondhand embarrassment (overly enthusiastic, trying too hard to be cool, excessive emoji use)
**Humblebragginess** - Disguising boasts as casual updates or false modesty ("just happened to close our $ 10M round 🤷")
**Vagueness** - Failing to clearly communicate what the product/service actually does
For each criterion, provide:
1. A score (1-5) where 1 = not present, 5 = severely present
2. If the problem is not present, no comment. Otherwise, explain of the issue and give concise guidance on fixing it, without providing an actual rewrite
@draft_tweet
"""
[pipe.optimize_tweet]
PipeLLM = "Optimize the tweet based on the analysis"
inputs = { draft_tweet = "DraftTweet", tweet_analysis = "TweetAnalysis", writing_style = "WritingStyle" }
output = "OptimizedTweet"
llm = "llm_for_social_post_writing"
system_prompt = """
You are an expert in writing engaging tech tweets that drive meaningful discussions and engagement.
Your goal is to rewrite tweets to be impactful and avoid the pitfalls identified in the analysis.
"""
prompt_template = """
Rewrite this tech tweet to be more engaging and effective, based on the analysis:
Original tweet:
@draft_tweet
Analysis:
@tweet_analysis
Requirements:
- Include a clear call-to-action
- Make it engaging and shareable
- Use clear, concise language
### Reference style example
@writing_style
### Additional style instructions
No hashtags.
Minimal emojis.
Keep the core meaning of the original tweet.
"""
[pipe.optimize_tweet_sequence]
PipeSequence = "Analyze and optimize a tech tweet in sequence"
inputs = { draft_tweet = "DraftTweet", writing_style = "WritingStyle" }
output = "OptimizedTweet"
steps = [
{ pipe = "analyze_tweet", result = "tweet_analysis" },
{ pipe = "optimize_tweet", result = "optimized_tweet" },
]Here is the flowchart generated during this run:
---
config:
layout: dagre
theme: base
---
flowchart LR
subgraph "optimize_tweet_sequence"
direction LR
FGunn["draft_tweet:<br>**Draft tweet**"]
EWhtJ["tweet_analysis:<br>**Tweet analysis**"]
65Eb2["optimized_tweet:<br>**Optimized tweet**"]
i34D5["writing_style:<br>**Writing style**"]
end
class optimize_tweet_sequence sub_a;
classDef sub_a fill:#e6f5ff,color:#333,stroke:#333;
classDef sub_b fill:#fff5f7,color:#333,stroke:#333;
classDef sub_c fill:#f0fff0,color:#333,stroke:#333;
FGunn -- "Analyze tweet" ----> EWhtJ
FGunn -- "Optimize tweet" ----> 65Eb2
EWhtJ -- "Optimize tweet" ----> 65Eb2
i34D5 -- "Optimize tweet" ----> 65Eb2
We welcome contributions! Please see our Contributing Guidelines for details on how to get started, including development setup and testing information.
Join our vibrant Discord community to connect with other developers, share your experiences, and get help with your Pipelex projects!
- GitHub Issues: For bug reports and feature requests
- Discussions: For questions and community discussions
- Documentation
If you find Pipelex helpful, please consider giving us a star! It helps us reach more developers and continue improving the tool.
This project is licensed under the MIT license. Runtime dependencies are distributed under their own licenses via PyPI.
"Pipelex" is a trademark of Evotis S.A.S.
© 2025 Evotis S.A.S.