Pipulate: Local First AI SEO Software
Project description
Pipulate: Local First AI SEO Software
Your data. Your AI. Your machine. Your control.
No subscriptions, no vendor lock-in, no cloud costs.
๐ Quick Start for Impatient People
Want to skip the philosophy and just see what this does?
# 1. Install Nix (one-time setup)
curl --proto '=https' --tlsv1.2 -sSf -L https://install.determinate.systems/nix | sh -s -- install
# 2. Close and reopen your terminal, then:
curl -L https://pipulate.com/install.sh | sh
# 3. Launch it
cd ~/pipulate && nix develop
What you get: A local web app at http://localhost:5001
with step-by-step workflows, integrated AI chat, and a JupyterLab instance at http://localhost:8888
. No cloud required.
Success looks like: Two browser tabs auto-open showing the Pipulate interface and JupyterLab.
๐ก What Can You Actually Build?
Real examples of what people create with Pipulate:
๐ SEO Workflows
- Keyword Research Pipeline: Input seed keywords โ AI expansion โ competition analysis โ export spreadsheet
- Content Gap Analysis: Compare your site vs competitors โ identify missing topics โ prioritized content calendar
- Technical SEO Audits: Crawl site โ check Core Web Vitals โ generate action items โ track fixes
๐ Data Processing Workflows
- CSV Data Cleaning: Upload messy data โ standardize formats โ remove duplicates โ validate results
- API Data Collection: Connect to APIs โ fetch data in batches โ transform to consistent format โ store locally
- Report Generation: Combine multiple data sources โ apply business rules โ create branded reports
๐ค AI-Assisted Workflows
- Content Creation Pipeline: Research topics โ generate outlines โ write drafts โ optimize for SEO
- Data Analysis Helper: Upload spreadsheet โ AI suggests insights โ create visualizations โ export findings
Key advantage: Each workflow is a guided, step-by-step process that non-technical users can run repeatedly, while developers can customize the Python code behind each step.
Meet Chip O'Theseus
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ Chip O'What?
โ ๐ญ PIPULATE: LOCAL-FIRST AI SEO SOFTWARE & DIGITAL WORKSHOP โ , O
โ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โ \\ . O
โ โ |\\/| o
โ ๐ฌ Chip O'Theseus: "Welcome to your sovereign computing environment!" โ / " '\
โ โ . . .
โ ๐ Where Python functions become HTML elements... โ / ) |
โ ๐ Where workflows preserve your creative process... โ ' _.' |
โ ๐ Where AI integrates locally and globally... โ '-'/ \
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
AI On Rails: Structured Workflows for Any AI
The Challenge with Agentic AI: Powerful but unpredictableโyou never know what you're gonna get.
The Pipulate Approach: Structured workflows that can leverage any AIโlocal, cloud, or hybridโwhile maintaining complete visibility and control.
Think of it as putting guardrails on AI assistance. Instead of asking an AI to "figure it out," domain experts create step-by-step workflows that guide AI through proven processes. The AI gets structure, you get predictable results.
Pipulate: Your AI Swiss Army Knife: Whether you prefer local privacy, cloud power, or hybrid approaches, Pipulate provides the framework. Use local models for sensitive work, cloud APIs for heavy lifting, or both in the same workflowโyour choice, your control.
๐ค AGENTIC MODE (Chaos) ๐ AI ON RAILS (Pipulate)
โโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโ
๐ฅ GOES OFF ๐ LINEAR WORKFLOWS
HALF-COCKED! BY DOMAIN EXPERTS
โ โ
โผ โผ
โโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโ
โ ๐ช๏ธ WILLY NILLY ๐ฒ โ โ Step 1: Analyzeโธ โ
โ โ VS โ Step 2: Processโธ โ
โ Unpredictable โ โ Step 3: Reportโธ โ
โ Results โ โ Step 4: Exportโธ โ
โโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโ
โ โ
โผ โผ
โ๏ธ Trains Frontier Models ๐ Keeps Domain Expertise Local
- ๐ฅ๏ธ Runs locally like a desktop app using modern web technologies
- ๐ Simple linear workflow approach powered by HTMX for seamless interactivity
- ๐ Transforms Jupyter Notebooks into production-ready, step-by-step workflows
- ๐ค Integrated AI assistance using your own local models with complete privacy
- ๐ง Reproducible environments with Nix that work identically across all platforms
- ๐ฏ Perfect for SEO practitioners who want to turn technical expertise into guided, reusable workflows
What is Pipulate?
Pipulate is a local-first, single-tenant desktop app framework featuring AI-assisted, step-by-step workflows. Designed to feel like an Electron app, it uniquely runs a full, reproducible Linux environment within a project folder using Nix, ensuring consistency across macOS, Linux, and Windows (via WSL).
Desktop App Architecture: Electron vs Pipulate
๐ฅ๏ธ ELECTRON PATTERN ๐ PIPULATE PATTERN
โโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโ
โโโโโโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ ELECTRON APP โ โ PIPULATE SETUP โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโค โโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ โโโโโโโ โโโโโโโ โโโโโโโ โ โ โโโโโโโโโโโโโโโโโโโโโโโ โ
โ โ.exe โ โ.dmg โ โ.deb โ โ โ โ install.sh โ โ
โ โโโโโโโ โโโโโโโ โโโโโโโ โ โ โ (Works on ALL OSes) โ โ
โ Per-OS Installers โ โ โโโโโโโโโโโโโโโโโโโโโโโ โ
โโโโโโโโโโโโโฌโโโโโโโโโโโโโโ โโโโโโโโโโโโโฌโโโโโโโโโโโโโโ
โ โ
โผ โผ
โโโโโโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโโ This is new.
โ ๐ฑ Native Window โ โ ๐ฅ๏ธ Terminal Console โ , O
โ โโโโโโโโโโโโโโโโโโโ โ โ โโโโโโโโโโโโโโโโโโโโโโโ โ \\ . O
โ โ Web Browser โ โ โ โ nix develop โ โ |\\/| o
โ โ (Bundled) โ โ โ โ Starting servers... โ โ / " '\
โ โ โโโโโโโโโโโโโ โ โ โ โ โ JupyterLab ready โ โ . . .
โ โ โ โ โ โ โ โ โ Pipulate ready โ โ / ) |
โ โ โ HTML โ โ โ โ โโโโโโโโโโโโโโโโโโโโโโโ โ ' _.' |
โ โ โ CSS โ โ โ + โโโโโโโโโโโโโโโโโโโโโโโโโโโ '-'/ \
โ โ โ JS โ โ โ โ
โ โ โ โ โ โ โผ
โ โ โโโโโโโโโโโโโ โ โ โโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ โโโโโโโโโโโโโโโโโโโ โ โ ๐ Regular Browser โ
โ โ โ โ โโโโโโโโโโโโโโโโโโโโโโโ โ
โ โผ โ โ โ localhost:5001 โ โ
โ โโโโโโโโโโโโโโโโโโโ โ โ โ โโโโโโโโโโโโโโโโโโโ โ โ
โ โ Node.js โ โ โ โ โ Python/HTMX โ โ โ
โ โ Runtime โ โ โ โ โ Workflows โ โ โ
โ โโโโโโโโโโโโโโโโโโโ โ โ โ โ Local AI โ โ โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโ โ โ โโโโโโโโโโโโโโโโโโโ โ โ
โ โโโโโโโโโโโโโโโโโโโโโโโ โ
โ
Feels like native app โโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ Multiple installers needed
โ Platform-specific builds โ
Universal installer
โ Update distribution complexity โ
Auto-updates via git
โ
Same experience all OSes
โ
Complete reproducibility
The Magnum Opus: Computing Sovereignty
This isn't just another framework โ it's a deliberate culmination of decades of tech evolution insights. Pipulate represents the "third act" approach to development (3rd time's the charm): choosing the most durable and lovable parts of the modern tech stack while rejecting the exhausting hamster wheel of framework churn.
If you are not an Empire builder and prefer craftsmanship over the rat race and want to build tools that last, then Pipulate may be for you. Pipulate embodies that philosophy โ maximum creative freedom with minimum technical debt, recapturing that old Webmaster feeling.
Core Philosophy: Local-First, WET, and AI-Augmented
Breaking Free: Durable Foundations for Any Approach
๐ก THE FRAMEWORK CHURN CYCLE ๐ฐ COMPUTING SOVEREIGNTY
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโโ
React โ Vue โ Angular โ Svelte ๐ฝ Your Hardware
โ โ ๐ฝ Your Data
Webpack โ Next.js โ Vite โ Remix ๐ฝ Your AI Choice
โ โ ๐ฝ Your Code
Docker โ K8s โ Cloud โ Serverless ๐ฝ Your Schedule
๐ตโ๐ซ Endless Learning ๐ฝ Your Hardware
๐ธ Migration Fatigue ๐ฝ Your Data
๐ Platform Lock-in ๐ฝ Your AI Choice
๐ Growing Complexity ๐ฝ Your Code
๐ฝ Your Schedule
WITH
โจ Durable Tools:
๐โโ๏ธ JUMP OFF THE WHEEL โข Python (30+ years)
โ โข SQLite (built-in)
โโโโโโโโโโโโโโโ โข HTML/HTTP (timeless)
โ PIPULATE โ โข Nix (reproducible)
โ Local-First โ โข Cloud APIs (by choice)
โ+ Any Cloud โ
โโโโโโโโโโโโโโโ ๐ฏ Third Act Philosophy:
"Choose tools that will
outlast any framework"
-
Local-First Sovereignty: Your data, code, and AI run on your hardware by defaultโextending to cloud services when you choose. This guarantees privacy, eliminates surprise costs, and gives you complete control over when and how to scale.
-
WET Workflows, DRY Framework: Workflows are intentionally "WET" (explicit & step-by-step) for maximum clarity and customizabilityโperfectly mirroring Jupyter Notebooks. The underlying framework is "DRY" for efficiency.
-
The AI Advantage: AI makes WET practical. Tedious code maintenance and refactoring, once a weakness of WET, is now an area where AI excels, turning repetition into a strength for rapid, context-aware development. Our breakthrough Workflow Reconstruction System exemplifies this: intelligent AST-based transplantation of workflow components eliminates traditional OOP inheritance complexity while maintaining perfect code precision.
-
Radical Transparency ("Know EVERYTHING!"): We reject opaque enterprise patterns in favor of complete observability. State is managed in transparent SQLite tables and JSON blobs, making the entire system intuitive and debuggable. No black boxes, ever.
-
Reproducibility with Nix: Nix Flakes provide a perfect, reproducible Linux environment on macOS, Linux, and Windows (WSL), solving the "works on my machine" problem.
-
Future-Proof Stack: We rely on durable standards: Python, SQLite, HTML, and HTMX. This is a framework built to last.
Primary Goals
- Empower End-Users (e.g., SEO Practitioners): Enable non-programmers to run powerful, AI-guided workflows (often ported from Jupyter Notebooks) without needing to interact with Python code directly.
- Serve Developers: Provide a simple, reproducible environment for building these workflows, leveraging integrated tooling like Jupyter, local LLMs, and a streamlined web framework.
The Technical Stack: Simple Yet Powerful
Pipulate's WET philosophy extends to its technology choices, favoring simple, durable tools over complex abstractions:
Not On My Machine Problem Fixed
The Cloud's popularity has been driven in part by developers not wanting to maintain multiple codebases or installers per OS. Thanks to Nix, that's all fixed.
- Nix Flakes: Manages dependencies and creates reproducible environments, ensuring consistency across developers and operating systems, with optional CUDA support. E.g. Is this a Linux-thing you're reading about here? A Windows thing? A Mac thing? The answer is: YES!!! All of the above โ and if you've got cool acceleration hardware, it will even take advantage and utilize that too. Best of all worlds.
____ _ _ .--. ___________
| _ \ __ _ _ ____ _(_)_ __ (_)_ __ ,--./,-. |o_o | | | |
| | | |/ _` | '__\ \ /\ / / | '_ \| \ \/ / / # \ |:_/ | | | |
| |_| | (_| | | \ V V /| | | | | |> < | | // \ \ |_____|_____|
|____/ \__,_|_| \_/\_/ |_|_| |_|_/_/\_\ \ / (| | ) | | |
`._,._,' /'\_ _/`\ | | |
Solving the "Not on my machine" problem well. \___)=(___/ |_____|_____|
Nix serves as the "Noah's Ark" โ preserving this perfect focus in a reproducible environment that works identically across all platforms. Once you've locked in the focus, it lasts for years or decades, all bottled up in infrastructure-as-code.
Other Key Technologies Used
Pipulate integrates a carefully selected set of tools aligned with its philosophy:
- FastHTML: A Python web framework prioritizing simplicity. It generates HTML directly from Python objects (no template language like Jinja2) and minimizes JavaScript by design, working closely with HTMX. It's distinct from API-focused frameworks like FastAPI. The Python function-naming is the HTML-template language.
The New LAMP Stack: Evolution in Simplicity
๐๏ธ ORIGINAL LAMP STACK (2000s) ๐ NEW LAMP STACK (2025)
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโโ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ ๐ง L: Linux โ โ ๐ง L: Linux + Nix โ
โ Single OS, manual setup โ โ Reproducible everywhere โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ ๐ A: Apache โ โ โก A: ASGI โ
โ Static config, restarts โ โ Async, hot reload โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ ๐๏ธ M: MySQL โ โ ๐ M: MiniDataAPI โ
โ Complex queries, joins โ โ Python-native simplicityโ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ ๐ง P: PHP โ โ ๐ P: Python + FastHTML โ
โ Mix of HTML/logic โ โ + HTMX โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ โ
โผ โผ
โโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ ๐ข Enterprise โ โ ๐ Local-First โ
โ Complexity โ โ Sovereignty โ
โ โ โ โ
โ โข Multi-server โ โ โข Single machine โ
โ โข Load balancers โ โ โข Integrated AI โ
โ โข Database clusters โ VS โ โข SQLite simplicity โ
โ โข DevOps overhead โ โ โข Nix reproducibility โ
โ โข Cloud lock-in โ โ โข Flexible deployment โ
โโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโโ
๐ฏ One person understands ๐ฏ One person controls
part of the system the entire system
The original LAMP stack was beautiful in its simplicity โ one person could understand and manage the whole stack. But it got bloated with enterprise patterns, microservices, and distributed complexity.
Pipulate brings back that "one person, full stack" philosophy with modern tools:
- Linux + Nix: Reproducible environments across all platforms
- ASGI: Modern async server interface, future-proofed for performance
- MiniDataAPI: Universal SQL simplifier close to Python's core data structures
- Python + FastHTML + HTMX: The new web development paradigm
This isn't just simpler โ it's more powerful, giving you complete environment reproducibility, local AI integration, server-side state management, and future-proofed skills.
The Lens Stack: Focused Architecture
Pipulate's technology choices form aligned lenses that focus ideas from abstraction to actualization. Each lens must be ground and polished without misaligning the focus:
Universal Translator of Abstractions clarify into implementations
Spoken Language to Code by each lens being simple and transparent.
Idea --> Lens 1 --> Lens 2 --> Lens 3 -> Lens 4 -> Lens 5 -> Lens 6
-----> ,--.
---> ,' `.---------> ,--.
--> / \------> ,' `.-------> ,--. ,-.
o -> / Linux \----> / http \----> ,'_hx `.--->,' `. ,-.
/|\ ( HARDWARE )--> ( PROTOCOL )--> ( LINGUA )->( UI/UX )->(APP)->(git)
/ \ -> \ Nix /----> \ html /----> `..py ,'--->`. ,' `-'
--> \ /------> `. ,'-------> `--' `-' And so on
---> `. ,'---------> `--' AI Help
-----> `--' AI Help
AI Help
We keep lenses minimal, their material either thoroughly pre-trained into the model (Python 3.x, HTMX, etc.) or able to be included in the prompt and easily held in the context window. We've trimmed the cruft โ the lens flashes and burrs, and all unnecessary extra lenses (Angular, React, Vue, etc.)
HARDWARE:
install.sh: Published on Pipulate.com to initiate magic cookie install
flake.nix: Nix IaC creating a normalized Linux subsystem on any host OS
PROTOCOL:
http: Uvicorn fast Asynchronous Server Gateway Interface (ASGI) web server
html: Uvicorn talks to Python Starlette using anyio & httpx libraries
websocket: static/ws.js provides client bi-directional asynchronous communication
LINGUA:
htmx: static/htmx.js JavaScript library to eliminate most need for JavaScript
Python: .venv/bin/python3.12 latest version AIs are well trained on
UI/UX:
browser: Obviously, but I guess it needs to be said. Like a looser Electron.
fasthtml: static/fasthtml.js for FT Components, Python functions as templating
APP:
app: Flask-style Uvicorn factory instance instantiated by FastHTML fast_app
db: Dict-like DB providing transparent server-side state (server cookies)
pipulate: Pipeline state management, like db but with JSON blob for workflows
Grinding Off the Burrs and Flashes
In lens manufacturing, "flashes" are excess material that squeeze out of molds โ unwanted projections that must be ground off. Steve Jobs famously did this twice: adopting Gorilla Glass (grinding off plastic flashes) and rejecting Flash Player (grinding off software bloat).
Pipulate continues this tradition:
- FastHTML: Grinds off Jinja2 template complexity
- HTMX: Grinds off virtual DOM overhead
- Local AI: Enables privacy by default, cloud power when desired
- SQLite: Grinds off enterprise database complexity
The result: clean, focused tools that do their job without unnecessary cruft.
From Flask to FastAPI to FastHTML
This is not your father's Python web framework. HTMX changes everything โ a marriage made in heaven between Python and the Web, finally turning Python into a first-class citizen for web development. In many use cases such as this one, Python is even preferable to JavaScript in the way it blends Python's formidable ecosystem of packages with workflows.
The Evolution: Flask โ FastAPI โ FastHTML
The revolution isn't just another framework โ it's eliminating the template layer entirely:
๐ถ FLASK ERA ๐ FASTAPI ERA ๐ FASTHTML ERA
โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโ
โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ
โ Python โ โ Python โ โ Python โ
โ Functions โ โ Functions โ โ Functions โ
โโโโโโโโฌโโโโโโโ โโโโโโโโฌโโโโโโโ โโโโโโโโฌโโโโโโโ
โ โ โ
โผ โผ โผ
โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ
โ Jinja2 โ โ Pydantic โ โ HTMX โโโ Over-the-wire
โ Templates โ โ Models โ โ Fragments โ HTML targeting
โโโโโโโโฌโโโโโโโ โโโโโโโโฌโโโโโโโ โโโโโโโโฌโโโโโโโ DOM elements
โ โ โ
โผ โผ โผ
โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ
โ HTML โ โ JSON โ โ HTML โ
โ Response โ โ Response โ โ Elements โ
โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ
โ โ โ
โผ โผ โผ
๐ Full Page Reload ๐ฑ Frontend Framework ๐ฏ DOM Element Updates
(React/Vue/Angular) def Div() = <div>
def Button() = <button>
Template files needed JSON โ HTML conversion Python functions ARE
Separate languages Client-side complexity the template language!
The FastHTML Breakthrough: Python function names directly become HTML elements, eliminating templates and making the server the single source of truth for UI state.
-
HTMX: Enables dynamic, interactive UIs directly in HTML via attributes, minimizing the need for custom JavaScript. Pipulate uses it for server-rendered HTML updates โ over the wire HTML-fragments targeting elements of the DOM directly instead of fragile, performance-reducing, framework-dependent JSON. THIS is where you jump off the tech-churn hamsterwheel and future-proof yourself.
-
MiniDataAPI: A lightweight layer for interacting with SQLite and other databases. Uses Python dictionaries for schema definition, promoting type safety without the complexity of traditional ORMs โ effectively future-proofing your SQL. You lose fancy join capabilities but in exchange get the Python dict interface as your main persistent database API forever-forward, enabiling instant swapability between SQLite and PostgreSQL (for example).
-
Ollama: Facilitates running LLMs locally, enabling in-app chat, workflow guidance, and future automation capabilities while ensuring privacy and avoiding API costs. Your local AI (Chip O'Theseus) learns & grows with you, hopping from hardware to hardware as you upgrade โ like a genie in a hermitcrab shell. And if that weren't kooky enough โ it knows how to make MCP-calls!!! That's right, your friendly localhost AI Chip O'Theseus is also an MCP client! Your linear workflows ain't so linear anymore when a single-step can be: "Go out and do whatever."
The Hybrid Advantage: Best of Both Worlds
Pipulate isn't anti-cloudโit's pro-choice. Each workflow step can choose the best tool for the job:
- Step 1: Use local AI for sensitive data analysis (privacy-first)
- Step 2: Call OpenAI's API for advanced reasoning (cloud power)
- Step 3: Process results locally and save to SQLite (data sovereignty)
- Step 4: Use Anthropic's API for final review (frontier capabilities)
This is the Swiss Army knife approach: Local by default, cloud by choice, with complete visibility into what's happening at each step. Whether you're processing confidential client data (local) or need cutting-edge AI capabilities (cloud), Pipulate gives you the framework to do both seamlessly.
- SQLite & Jupyter Notebooks: Foundational tools for data persistence and the workflow development process (porting from notebooks to Pipulate workflows). SQLite is built into Python and really all things โ the get-out-of-tech-liability free card you didn't know you had. And a full JupyterLab instance is installed side-by-side with Pipulate sharing the same Python
.venv
virtual environment, which is also shared with your preferred AI code editor (Cursor, Windsurf, VSCode, Zed) so... well... uhm, there are no words for when 3 different portals-to-Python share the same environment. You can do such stupid AI-tricks as letting your local LLM and a frontier cloud model inhabit the same body (Pipulate) โ controlling web browsers together and stuff.
How to Install Pipulate
Installation Strategy: Universal First, PyPI Alternative
We offer two installation paths that lead to the exact same robust, Nix-managed environment. Choose the path that best fits your experience level and preferences.
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ New User on macOS โ
โโโโโโโโโโโโโโโฌโโโโโโโโโโโโโโโ
โ
โโโโโโโโโโโโโโโโโโโโโโโโโดโโโโโโโโโโโโโโโโโโโโโโโโ
โ โ
โผ โผ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ PATH 1: Recommended for Everyone โ โ PATH 2: Alternative for Python Developers โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ โ
"I want the simplest, most "I prefer managing my command-line
direct way to get this running." tools with standard Python utilities."
โ โ
โผ โผ
1. `curl ... [nix]` 1. `brew install pipx` (If needed)
2. `curl ... [pipulate]` 2. `pipx install pipulate`
3. `pipulate install`
โ โ
โโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโ
โ โ
โผ โผ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ Nix-Managed Pipulate โ
โ Environment โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
||
(Identical
Result)
PATH 1: Quick Start โ Universal Installation (Recommended)
This is the fastest and most universal way to install Pipulate. It has the fewest dependencies and works on any modern Mac, Linux system, or Windows with WSL.
๐ฆ Your Machine ๐ง Add Foundation ๐ Complete Environment
Today with Nix Ready to Go!
โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ
โ Sad Computerโ Step 1 โ ๐๏ธ Nix โ Step 2 โ ๐ฏ Pipulate โ
โ Without โ โโโโโโโโโโโบ โ Foundation โ โโโโโโโโโโโบ โ + AI + โ
โ Nix๐ข โ โ Installed โ โ Jupyter โ
โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ
โ
Step 3 โ
โผ
โโโโโโโโโโโโโโโ
โ ๐ Browser โ
โ Opens โ
โAutomaticallyโ
โโโโโโโโโโโโโโโ
Simple as 1-2-3! No Docker, no build steps, works with or without cloud services.
Everything runs locally with complete flexibility and control.
Step 1: Install Nix (One-Time Setup)
If you don't have it already, install the Nix package manager. It's the system that makes Pipulate's reproducible environment possible.
curl --proto '=https' --tlsv1.2 -sSf -L https://install.determinate.systems/nix | sh -s -- install
Important: After the Nix installation finishes, you must close and reopen your Terminal window.
Step 2: Run the Pipulate Installer
Now, run the universal install script. You can give your project a custom name, too.
# To install with a custom name like "Botifython"
curl -L https://pipulate.com/install.sh | sh -s Botifython
# Or, to install with the default name "pipulate"
curl -L https://pipulate.com/install.sh | sh
Step 3: Launch Pipulate
Navigate into your new project directory and launch the environment with nix develop
.
# cd into the directory you just created
cd ~/Botifython
# Launch Pipulate
nix develop
That's it! The server and JupyterLab will start, and the application will open in your browser.
Running It Again:
- You can just forcibly exit out of that Terminal it's running from.
- Open a new Terminal, and once again:
cd ~/Botifython
nix develop
The Big Reset (If Necessary):
Things sometimes go wrong. This is how you do a full Pipulate reset. This will also delete anything you downloaded with Pipulate. Adjust custom install name to what you used.
rm -rf ~/Botifython
curl -L https://pipulate.com/install.sh | sh -s Botifython
cd ~/Botifython
nix develop
Wait for BOTH TABS to auto-open in your browser.
๐จ Installation Troubleshooting
Common Issues & Solutions:
Problem | Solution |
---|---|
nix: command not found |
You didn't restart your terminal after Nix installation |
Browser doesn't open automatically | Manually visit http://localhost:5001 and http://localhost:8888 |
Permission denied errors |
Make sure you can write to ~/pipulate directory |
Port conflicts | Kill processes on ports 5001/8888: lsof -ti:5001 | xargs kill -9 |
Nix build fails | Clear Nix cache: nix-collect-garbage then retry |
System Requirements:
- macOS: 10.15+ (Intel/Apple Silicon)
- Linux: Any modern distribution with curl
- Windows: WSL2 with Ubuntu 20.04+
- RAM: 4GB minimum, 8GB recommended
- Disk: 2GB for installation + data storage
- Network: Internet connection for initial setup only
PATH 2: Alternative Installation via PyPI (For Python Developers)
If you are a developer comfortable with tools like Homebrew and pipx
, you can use our PyPI package as a gateway to the same robust installation.
Step 1: Install pipx
pipx
is a tool for safely installing Python command-line applications. If you don't have it, you can install it with Homebrew.
brew install pipx
Step 2: Install the Pipulate CLI
Use pipx
to install the pipulate
command-line tool. This will not cause conflicts with your system Python.
pipx install pipulate
Step 3: Run the Installer
Use the command you just installed to set up the full Pipulate application.
pipulate install
This will trigger the same universal installation process, resulting in the exact same robust, Nix-managed environment. To run it in the future, just type pipulate run
.
These few commands:
- โ Updates to the latest version automatically
- โ Starts JupyterLab and the Pipulate server
- โ Opens web interfaces in your browser
- โ Provides a complete, reproducible development environment
That's it! You now have a local-first development environment with AI integration, installed via your preferred Python toolchain.
Installation Process Deep Dive
Here's what happens behind the scenes during the "magic cookie" installation:
User runs install.sh (via curl) Nix Flake Activation & Transformation
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ 1. Download install.sh โ โ 5. User runs 'nix develop' โ
โ 2. Download ZIP from GitHub โ โ 6. Flake detects non-git directory โ
โ 3. Extract ZIP to ~/AppName โ โ 7. Flake clones repo to temp dir โ
โ 4. Download ROT13 SSH key โ โ 8. Preserves app_name.txt, .ssh, .venv โ
โ to .ssh/rot โ โ 9. Moves git repo into place โ
โโโโโโโโโโโโโโโฌโโโโโโโโโโโโโโโโโ โ10. Sets up SSH key for git โ
โ โ11. Transforms into git repo โ
โผ โ12. Enables auto-update via git pull โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ Result: Fully functional, auto-updating, git-based Pipulate installation โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
Chef or Customer?
Are you a Developer or an End User? Chef or Customer? Understanding your audience is crucial for effective development. Pipulate serves two distinct but complementary audiences, much like a restaurant serves both chefs and customers
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ The Restaurant โ
โ โโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโ โ
โ โ Kitchen (Dev) โ โ Dining Room โ โ
โ โ โ โ (End Users) โ โ
โ โ โ โ โ โ
โ โ ๐จโ๐ณ Sous Chef โโโโrecipesโโโโบโ ๐ฝ๏ธ Customers โ โ
โ โ ๐ฉโ๐ณ Head Chef โ โ ๐ข Restaurateur โ โ
โ โ โ โ โ โ
โ โ "How do we make โ โ "I want the best โ โ
โ โ pasta you've โ โ pasta I've ever โ โ
โ โ never had?" โ โ had in my life" โ โ
โ โโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโ โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
๐จโ๐ณ The Chef (Developer/Technical User)
- ๐ง Workflow Creators: Build and customize AI-assisted workflows
- ๐ Jupyter Porters: Convert notebook experiments into guided applications
- ๐ Technical SEOs: Create sophisticated, reusable SEO processes
- โ๏ธ System Administrators: Deploy consistent environments across teams
What Chefs Get:
- ๐๏ธ Complete control over the "recipe" (workflow logic)
- ๐ Reproducible development environment via Nix
- ๐๏ธ Simple architecture that's easy to understand and modify
- ๐งฐ Integrated tooling (Jupyter, local LLM, SQLite)
๐ฝ๏ธ The Customer (End User/Non-Technical)
- ๐ SEO Practitioners: Run powerful workflows without coding
- โ๏ธ Content Creators: Use AI-assisted processes for optimization
- ๐ Marketing Teams: Execute consistent SEO strategies
- ๐ข Business Owners: Access enterprise-level SEO capabilities
What Customers Get:
- ๐ถโโ๏ธ Guided, step-by-step workflow experiences
- ๐ค AI assistance at every step
- ๐ No need to see or understand the underlying code
- ๐ฏ Consistent, repeatable results
๐ The Restaurant Analogy
Just as a chef talks about knife techniques while a diner just wants amazing pasta, Pipulate separates the complexity of creation from the simplicity of consumption. Developers craft the workflows, end-users enjoy the results.
๐ฏ Your First 10 Minutes with Pipulate
After installation succeeds, here's what to expect:
What You'll See
-
Two browser tabs open automatically:
localhost:5001
- Pipulate web interface with navigation menulocalhost:8888
- JupyterLab for development/experimentation
-
In the Pipulate interface:
- Left sidebar with workflow plugins (Introduction, Profiles, etc.)
- Main area showing step-by-step workflow interface
- Right panel with integrated AI chat (Chip O'Theseus)
-
Terminal shows:
๐ Starting Pipulate servers... โ FastHTML server ready at http://localhost:5001 โ JupyterLab ready at http://localhost:8888 โ Local AI ready for chat assistance
Your Next Steps Depend on Who You Are
๐ If you're an SEO practitioner:
- Click "Introduction" in the left menu for a guided tour
- Try the built-in workflows to see the step-by-step pattern
- Use the AI chat to ask "How do I create a keyword research workflow?"
๐จโ๐ป If you're a developer:
- Open JupyterLab tab and run the introduction notebook
- Check out
plugins/010_introduction.py
to see workflow code structure - Try creating a simple workflow:
python helpers/workflow/create_workflow.py
๐ค If you're an AI assistant:
- Focus on the Quick Reference Card above
- Study the Critical Implementation Patterns section
- Review
mcp_tools.py
for MCP protocol capabilities
๐ If you're just exploring:
- Click through the left menu items to see different workflow types
- Ask the AI chat: "What can I build with Pipulate?"
- Try the Introduction workflow to see the step-by-step experience
The WET Revolution: Why Explicit Code Wins in the AI Era
Pipulate is built on a radical philosophy that challenges programming orthodoxy: WET (Write Everything Twice) is better than DRY (Don't Repeat Yourself) when you have AI to help manage it.
The Universal API Pattern: From Quarks to Code
At every scale of reality, we see the same pattern: "lumps of stuff" with APIs that enable interaction. Quarks combine into atoms, atoms into molecules, cells into organisms, individuals into societies. Each level requires the right granularity of interface โ not so abstract that you lose control, not so granular that you drown in complexity.
This is the 80/20 rule of existence: Handle 80% of interactions gracefully with 20% of the API surface, then handle edge cases as needed. Pipulate applies this principle to code architecture.
Durable vs. Ephemeral: Building on Bedrock
The tech industry suffers from "hamster wheel syndrome" โ constantly breaking APIs that force migration cycles. React (20+ versions), Node (frequent breaking changes), Angular (complete rewrites). This isn't progress; it's planned obsolescence.
Pipulate chooses durable foundations:
- Linux Kernel: Version 6 in 30 years
- Python: Version 3 in 30 years
- HTML: Version 5 and stable
- HTTP: Version 3 and backward compatible
These are the "laws of physics" for software โ stable APIs that enable compound growth rather than constant rebuilding.
Why WET Works Now
Traditional development follows DRY principles, creating abstract, complex systems that are hard to understand and modify. But the world has changed:
- ๐ฌ Jupyter Notebooks promote explicit, literate programming
- ๐ค AI assistants excel at managing repetitive code
- ๐ Local-first architectures prioritize clarity over enterprise complexity
- ๐ฏ Right Granularity: WET provides the perfect abstraction level for human AND AI comprehension
________________________________
- Like Notebooks / \
- Linear Workflows | It runs proprietary private AI |
- Local & Cloud-free | Workflows from your Local PC?! |
- Chip O'Theseus included \________________________________/
()
HARDWARE PLATFORM LOCAL BROWSER O , Chip O'Theseus
_______________________ __________ _______ o \\ .
| | / Pipulate \Jupyter\__ |\\/|
| Windows, Mac or Linux | | __________________ | See! / " '\ - Radical transparency
| _____ ___ | | | App Name Menuโ๏ธ| |<- - - - -. . . - MCP tool-call control
| _/ Nix \____\_____ | | |------------------| | / ) | - Browser as bot's body
| | | | | | Workflow | Local | | ' _.' |
| | Pipulate <---------> -Step #1 | AI๐ค | | '-'/ \
|__| localhost:5001 |_| | | -Step #2 | Chat | | What, no Docker?
| (AI on Rails๐) | | | -Step #3 | Helpโธ | | What, no React?
|__________________| | |__________|_______| | What, no Cloud?
|______________________|
WET workflows are:
- ๐ Observable: See exactly what's happening at every step
- ๐ง Customizable: Modify workflows without breaking abstractions
- ๐ค AI-Friendly: Clear code that AI assistants can easily understand and maintain
- ๐ Future-Proof: Built on durable web standards that won't become obsolete
Developer Setup & Environment Notes
Nix Environment Activation: Always run nix develop
from the ~/pipulate
directory before running any project commands (python server.py
, pip install
, etc.) in a new terminal. This ensures you are using the correct dependencies defined in flake.nix
.
Interactive vs. Quiet Shell:
Standard Shell: nix develop
(or nix develop .#default
) runs the startup script (run-script
defined in flake.nix
) with welcome messages and service startup. Ideal for general use.
Quiet Shell: nix develop .#quiet
activates the Nix environment without running the full startup script or launching services automatically. It only sets up paths and installs pip requirements. Use this for:
- Running specific commands without starting the servers (e.g.,
nix develop .#quiet --command python -c "import pandas"
). - Debugging or interacting with AI assistants where verbose startup output is undesirable.
- Manually running
run-server
orrun-jupyter
(scripts placed in.venv/bin
by theshellHook
).
Dependencies: System-level dependencies (Python version, libraries like gcc
, zlib
) are managed by flake.nix
. Python package dependencies are managed by pip
using requirements.txt
within the Nix-provided environment.
Source of Truth: The flake.nix
file is the definitive source for the development environment setup.
Architecture & Key Concepts
Pipulate features a distinct architecture designed for its local-first, simple, and observable nature.
Architecture Overview Diagram
This diagram illustrates the high-level components and their interactions:
โโโโโโโโโโโโโโโ Like Electron, but full Linux subsystem
โ Browser โ in a folder for macOS and Windows (WSL)
โโโโโโโฌโโโโโโโโ
โ HTTP/WS
โผ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ Nix Flake Shell โ - In-app LLM (where it belongs)
โ โโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโ โ - 100% reproducible
โ โ FastHTML โ โ Ollama โ โ - 100% local
โ โ HTMX App โ โ Local LLM โ โ - 100% multi-OS
โ โโโโโโโโโฌโโโโโโโโ โโโโโโโโโโโโโโโโ โ
โ โ โ
โ โโโโโโโผโโโโโโ โโโโโโโโโโโโโโ โ
โ โMiniDataAPIโโโโโโบโ SQLite DB โ โ
โ โโโโโโโโโโโโโ โโโโโโโโโโโโโโ โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
This complete, self-contained environment runs identically on any operating system, providing the foundation for all Pipulate workflows and AI interactions.
Integrated Data Science Environment
Jupyter Notebooks run alongside the FastHTML server, allowing developers to prototype workflows in a familiar environment before porting them to Pipulate's step-based interface for end-users. The same Python virtual environment (.venv
) is shared, and ad-hoc package installation is supported. If you're using Cursor, VSCode or Windsurf, set your Ctrl
+Shift
+P
"Python: Set Interpreter" to "Enter Interpreter Path" ./pipulate/.venv/bin/python
. You might have to adjust based on the folder you use as your workspace. But then you'll have a Python environment unified between Cursor, JupyterLab and Pipulate.
โโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโ
โ Jupyter Lab โ โ FastHTML โ
โ Notebooks โ โ Server โ
โ โโโโโโโโโโโโ โ โ โโโโโโโโโโโโ โ
โ โ Cell 1 โ โ โ โ Step 1 โ โ
โ โ โ โ--->โ โ โ โ
โ โโโโโโโโโโโโ โ โ โโโโโโโโโโโโ โ
โ โโโโโโโโโโโโ โ โ โโโโโโโโโโโโ โ
โ โ Cell 2 โ โ โ โ Step 2 โ โ
โ โ โ โ--->โ โ โ โ
โ โโโโโโโโโโโโ โ โ โโโโโโโโโโโโ โ
โ localhost:8888 โ โ localhost:5001 โ
โโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโ
Local-First & Single-Tenant Details
Pipulate manages all state server-side within the local environment (think local-server cookies), with optional cloud integration as needed. This approach offers:
- Privacy & Control: Data stays local by default, cloud integration when you choose.
- Full Resource Access: Utilize local CPU/GPU freely for intensive tasks, plus cloud APIs for heavy lifting.
- Simplicity: Eliminates complexities of multi-tenancy while supporting both local and cloud workflows.
- Observability: State changes (via DictLikeDB/JSON) are transparent and easily logged (AI greps it there).
Local-First State Management Benefits
This detailed view shows how Pipulate's local-first architecture eliminates common web development complexities:
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ # Benefits of Local-First Simplicity
โ Web Browser โ
โ โ - No mysterious client-side state
โ โโโโโโโโโโโโโโโโโโโโโโ โ - No full-stack framework churn
โ โ Server Console โ โ - No complex ORM or SQL layers
โ โ & Web Logs โ โ - No external message queues
โ โโโโโโโโโโโฌโโโโโโโโโโโ โ - No build step required
โ โผ โ - Direct, observable state changes
โ โโโโโโโโโโโโโโโโโโโโโโโ โ
โ โ Server-Side State โ โ - Conceptually like local-server-side cookies
โ โ DictLikeDB + JSON โโโโโโโโโ Enables the "Know EVERYTHING!" philosophy
โ โโโโโโโโโโโโโโโโโโโโโโโ โ - AI greps logs/server.log to see app state!
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
Server-Rendered UI (HTMX)
The UI is constructed primarily with server-rendered HTML fragments delivered via HTMX. This minimizes client-side JavaScript complexity.
- FastHTML generates HTML components directly from Python.
- HTMX handles partial page updates based on user interactions, requesting new HTML snippets from the server.
- WebSockets and Server-Sent Events (SSE) provide real-time updates (e.g., for chat, live development reloading).
HTMX+Python enables a world-class
Python front-end Web Development environment.
โโโโโโโโโโโโโโโโโโโโโโโ
โ Navigation Bar โ - No template language (like Jinja2)
โโโโโโโโโโโฌโโโโโโโโโโโโค - HTML elements are Python functions
Simple Python back-end โ Main โ Chat โ - Minimal custom JavaScript / CSS
HTMX "paints" HTML into โ Area โ Interface โ - No React/Vue/Angular overhead
the DOM on demand โโโโโโโโบ โ โ โ - No "build" process like Svelte
โโโโโโโโโโโดโโโโโโโโโโโโ - No virtual DOM, JSX, Redux, etc.
With such minimal surface area the AI code assistant knows everything. LLMs are either pre-trained on the stable, infrequently revved libraries used (Python 3.12, HTMX, or it's all small enough to fit in a 1-shot prompt โ yes, the whole core code-base fits in one Gemini Web UI form submit.
Workflow Patterns & Development
Pipeline Workflows
Designed for porting notebook-style processes, workflows are sequences of steps where the state is managed explicitly at each stage and stored persistently (typically as a JSON blob in the pipeline
table).
- Resumable & Interrupt-Safe: Because each step's completion is recorded, workflows can be stopped and resumed.
- Explicit State Flow: Data typically passes from one step's output (
done
field) to the next via thetransform
function, simplifying debugging. Patterned on Unix pipes. - Good Training Data: The structured input/output of each step creates valuable data for potentially fine-tuning models.
- Proprietary Friendly: Excellent for proprietary domain-experts and fields (competing academic, finances) who resist letting their data flow onto the Web for general AI training.
โโโโโโโโโโโ โโโโโโโโโโโ โโโโโโโโโโโ - Fully customizable steps
โ Step 01 โโpipedโโบโ Step 02 โโpipedโโบโ Step 03 โ - Interruption-safe & resumable
โโโโโโโโโโโ โโโโโโโโโโโ โโโโโโโโโโโ - Easily ported from Notebooks
โ โ โ - One DB record per workflow run
โผ โผ โผ - Everything stays on your machine
State Saved State Saved Finalized? - Magnitudes simpler than celery
Run All Cells Pattern
The key insight: Pipulate workflows use a run_all_cells()
pattern that directly mirrors Jupyter's "Run All Cells" command. This creates an immediate mental model โ each workflow step is like a notebook cell, and the system automatically progresses through them top-to-bottom, just like running all cells in a notebook.
๐ JUPYTER NOTEBOOK ๐ PIPULATE WORKFLOW
โโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโ
[ ] Cell 1: Import data โโโโโโโโโโโโโโโโโโโโโโโ
โ โ Step 1: Data Input โ
โผ โโโโโโโโโโโโฌโโโโโโโโโโโ
[โถ] Cell 2: Process data โ hx_trigger="load"
โ โผ
โผ โโโโโโโโโโโโโโโโโโโโโโโ
[ ] Cell 3: Generate report โ Step 2: Processing โ
โ โโโโโโโโโโโโฌโโโโโโโโโโโ
โผ โ hx_trigger="load"
[ ] Cell 4: Export results โผ
โโโโโโโโโโโโโโโโโโโโโโโ
๐ฏ "Run All Cells" Button โโโโบ โ Step 3: Export โ
Executes top-to-bottom โโโโโโโโโโโโโโโโโโโโโโโ
Same mental model, same execution flow!
But with persistent state, a web UI and
not having to look at the Python code ๐ซ๐.
LLM Integration (Ollama)
Integration with a local Ollama instance provides AI capabilities without external API calls:
- Privacy: Prompts and responses stay local.
- Cost-Effective: No per-token charges; run continuously using local resources.
- Streaming Support: Real-time interaction via WebSockets.
- Bounded Context: Manages conversation history effectively.
- App State Awareness: Grepping your server log reveals full application state.
- Tool Calling: Local LLM is an MCP client with a growing list of abilities
- Workflow assistance
- Browser automation
- Debugging
โโโโโโโโโโโโโโโโโโโโ
โ Local Ollama โ - No API keys needed
โ Server โ - Completely private processing
โโโโโโโโโโฌโโโโโโโโโโ
โ
โ Streaming via WebSocket
โผ
โโโโโโโโโโโโโโโโโโโโ
โ Pipulate App โ - Monitors WS for MCP tool-call commands
โ(WebSocket Client)โ - Parses responses in real-time
โโโโโโโโโโฌโโโโโโโโโโ
โ
โ In-memory or DB backed
โผ
โโโโโโโโโโโโโโโโโโโโ
โ Bounded โ - Manages context window (~128k)
โ Chat History โ - Enables RAG / tool integration
โโโโโโโโโโโโโโโโโโโโ
Multi-OS & CUDA Support (Nix)
Nix Flakes ensure a consistent environment across Linux, macOS, and Windows (via WSL), optionally leveraging CUDA GPUs if detected.
โโโโโโโโโโโโโโโโโโโโ
โ Linux / macOS โ - Write code once, run anywhere
โ Windows (WSL) โ - Consistent dev environment via Nix
โโโโโโโโโโฌโโโโโโโโโโ - As if Homebrew but across all OSes
โ
โ Nix manages dependencies
โผ
โโโโโโโโโโโโโโโโโโโโ
โ CUDA Support โ - Auto-detects NVIDIA GPU w/ CUDA
โ (if present) โ - Uses GPU for LLM acceleration
โโโโโโโโโโโโโโโโโโโโ - Falls back to CPU if no CUDA
UI Layout
The application interface is organized into distinct areas:
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ Navigation โโโ Search, Profiles,
โโโโโโโโโโโโโโโโโฌโโโโโโโโโโโโโโค Apps, Settings
โ โ โ
Workflow, โโโบ Main Area โ Chat โ
App UI โ (Pipeline) โ Interface โโโ LLM Interaction
โ โ โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
UI Component Hierarchy: Complete DOM Structure with IDs & ARIA Labels
Critical for AI assistants: All UI components use semantic IDs and comprehensive ARIA labeling for accessibility and automation.
๐ home (Root Component)
โโโ ๐ฆ create_outer_container()
โ โโโ ๐งญ create_nav_group() [id='nav-group', role='navigation', aria-label='Main navigation']
โ โ โโโ ๐ nav_search_container [role='search', aria-label='Plugin search']
โ โ โ โโโ Input [id='nav-plugin-search', role='searchbox', aria-label='Search plugins']
โ โ โ โโโ Div [id='search-results-dropdown', role='listbox', aria-label='Search results']
โ โ โโโ ๐ค create_profile_menu() [id='profile-dropdown-menu', aria-label='Profile management']
โ โ โ โโโ Summary [id='profile-id', aria-label='Profile selection menu']
โ โ โ โโโ Ul [role='menu', aria-label='Profile options', aria-labelledby='profile-id']
โ โ โโโ โก create_app_menu() [id='app-dropdown-menu', aria-label='Application selection']
โ โ โ โโโ Summary [id='app-id', aria-label='Application menu']
โ โ โ โโโ Ul [role='menu', aria-label='Application options', aria-labelledby='app-id']
โ โ โโโ ๐ create_env_menu() [id='env-dropdown-menu', data-testid='environment-dropdown-menu']
โ โ โ โโโ Summary [id='env-id', aria-label='Environment selection menu']
โ โ โ โโโ Ul [role='menu', aria-label='Environment options', aria-labelledby='env-id']
โ โ โโโ โ๏ธ poke_section [id='poke-dropdown-menu']
โ โ โโโ Summary [id='poke-summary']
โ โ โโโ Div [id='nav-flyout-panel']
โ โโโ ๐ฑ main-grid
โ โ โโโ ๐ create_grid_left() [id='grid-left-content'] โ Workflow Steps/Cells Display
โ โ โ โโโ content_to_render (Dynamic workflow content)
โ โ โ โโโ scroll_to_top [id='scroll-to-top-link']
โ โ โโโ ๐ค create_chat_interface() [id='chat-interface', role='complementary', aria-label='AI Assistant Chat']
โ โ โโโ H2 [APP_NAME + ' Chatbot']
โ โ โโโ Div [id='msg-list', role='log', aria-label='Chat conversation', aria-live='polite']
โ โ โโโ Form [role='form', aria-label='Chat input form']
โ โ โโโ Textarea [id='msg', role='textbox', aria-label='Chat message input', aria-multiline='true']
โ โ โโโ Button [id='send-btn', aria-label='Send message to AI assistant']
โ โ โโโ Button [id='stop-btn', aria-label='Stop AI response streaming']
โ โโโ ๐ง HTMX Refresh Listeners
โ โโโ Div [id='profile-menu-refresh-listener', hx_target='#profile-dropdown-menu']
โ โโโ Div [id='app-menu-refresh-listener', hx_target='#app-dropdown-menu']
๐ฏ Key HTMX Targets for AI Browser Automation
Navigation Updates:
#profile-dropdown-menu
- Profile menu refresh target#app-dropdown-menu
- App menu refresh target#search-results-dropdown
- Live search results#nav-flyout-panel
- Settings flyout panel
Content Areas:
#grid-left-content
- Main workflow display area#msg-list
- Chat conversation historybody
- Full page navigation refreshes
Interactive Elements:
#nav-plugin-search
- Real-time plugin search (300ms delay)#send-btn
/#stop-btn
- Chat control buttons#scroll-to-top-link
- Scroll navigation aid
This structure enables AI assistants to programmatically interact with all UI components using semantic selectors and ARIA landmarks.
File Structure
.
โโโ .cursor/ # Bootstraps Radical Transparency (teaches AI to fish)
โ โโโ rules/ # Framework rules (01_CRITICAL_PATTERNS.mdc, etc.)
โโโ .venv/ # Common Python environment for FastHTML, Jupyter & Cursor
โโโ browser_automation/ # Selenium browser control & DOM capture
โ โโโ looking_at/ # Current browser DOM state for AI visibility
โ โโโ *.py # Google search automation examples
โโโ cli.py # Command line interface for Pipulate operations
โโโ common.py # Base Class for DRY CRUD plugin app inheritance (todo)
โโโ data/
โ โโโ data.db # AI-accessible SQLite for application state (server cookies)
โโโ downloads/ # Default location for workflow outputs (e.g., CSVs)
โโโ helpers/
ย ย โย ย โโโ botify
ย ย โย ย โย ย โโโ botify_api.ipynb # Git managed massive example notebook, produces docs
ย ย โย ย โโโ workflow # Workflow workshop, lots of tools that make WET DRY
ย ย โย ย โย ย โโโ create_workflow.py # Example of what might be found there
โ โโโ prompt_foo.py # Bundles XML code payloads for massive 1-shot AI prompts
โโโ logs/
ย ย โย ย โโโ server-1.log # N-rotations of server log per run per config
โ โโโ server.log # The server log of most recent run, contains app state
โโโ static/ # JS, CSS, images, icons
โโโ plugins/ # Workflow plugins (010_introduction.py, 400_trifecta.py, etc.)
โโโ pyproject.toml # Python packaging configuration and metadata
โโโ training/ # Markdown files for AI context/prompts
โโโ vulture_whitelist.py # Code analysis whitelist for unused code detection
โโโ flake.nix # Infrastructure as Code & all system-versions for AI
โโโ LICENSE # It's MIT
โโโ install.sh # "Magic cookie" installation script (curl | sh)
โโโ mcp_tools.py # MCP protocol tools - the AI assistant interface
โโโ notebook_introduction_local.ipynb # Editable (non-auto-updating) copy of botify_api.ipynb
โโโ README.md # This file
โโโ requirements.txt # Python dependencies (managed by Nix)
โโโ server.py # Main application entry point
Critical Implementation Patterns for LLMs
These patterns are essential for LLMs working with Pipulate and are frequently missed:
1. The Auto-Key Generation Pattern (MOST CRITICAL)
๐ AUTO-KEY GENERATION FLOW
โโโโโโโโโโโโโโโ POST โโโโโโโโโโโโโโโ HX-Refresh โโโโโโโโโโโโโโโ
โ Empty Form โ โโโโโโโโโโโบ โ Server โ โโโโโโโโโโโโโโโบ โ Page Reload โ
โ Submit โ โ /init โ Response โ Header โ Trigger โ
โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ
โฒ โ
โ โผ
โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ
โ User Hits โ โโโโโโโโโโโโ โ Auto-Key โ โโโโโโโโโโโโโโ โ landing() โ
โ Enter Again โ Ready! โ Populated โ Generates โ Method โ
โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโ
When a user hits Enter on an empty key field, this specific sequence occurs:
- Form Submission: POSTs to
/{APP_NAME}/init
with emptypipeline_id
- Server Response: The
init
method MUST return anHX-Refresh
response:if not user_input: from starlette.responses import Response response = Response('') response.headers['HX-Refresh'] = 'true' return response
- Page Reload: HTMX triggers a full page reload
- Auto-Key Population: The
landing()
method callspip.generate_pipeline_key(self)
to populate the input field - User Interaction: User hits Enter again to start the workflow
2. The Chain Reaction Pattern: The run_all_cells()
Breakthrough
Pipulate uses HTMX-driven step progression powered by the brilliantly named run_all_cells()
method:
- Initial Trigger: After
init
, therun_all_cells()
method initializes the workflow just like Jupyter's "Run All Cells" - Perfect Mental Model: The method name creates immediate understanding โ workflows execute top-to-bottom like notebook cells
- Step Handlers: Each step has GET (display) and POST (submit) handlers
- Automatic Progression: Completed steps trigger next step with
hx_trigger="load"
- State Persistence: Each step stores data in pipeline state
- Pedagogical Brilliance: The naming makes the system instantly intuitive for developers and AI assistants
Example: The run_all_cells()
Pattern in Action
# โ
CORRECT: Use the run_all_cells() method for workflow initialization
async def init(self, request):
"""Initialize workflow using the run_all_cells pattern"""
return pip.run_all_cells(app_name, steps)
# โ ANTI-PATTERN: Manual placeholder creation
async def init(self, request):
"""Manual approach โ harder to understand and maintain"""
first_step_id = steps[0].id
return Div(
Div(id=first_step_id, hx_get=f'/{app_name}/{first_step_id}', hx_trigger='load'),
id=f"{app_name}-container"
)
The run_all_cells()
method encapsulates the workflow initialization pattern and creates an immediate mental connection to Jupyter notebooks.
3. APP_NAME vs. Filename Distinction
๐ FILENAME vs APP_NAME DISTINCTION
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ CRITICAL SEPARATION โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ โ
โ ๐ FILENAME: 200_workflow_genesis.py โ
โ โโโ ๐ Determines public URL: /workflow_genesis โ
โ โโโ ๐ Controls menu order: 200 โ
โ โ
โ ๐ท๏ธ APP_NAME: "workflow_genesis_internal" โ
โ โโโ ๐พ Database table identifier โ
โ โโโ ๐ MUST REMAIN STABLE (data integrity) โ
โ โโโ ๐ซ NEVER change after deployment โ
โ โ
โ โ ๏ธ DANGER: Changing APP_NAME = Orphaned Data โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
Critical for data integrity:
- Filename (e.g.,
200_workflow_genesis.py
): Determines public URL endpoint and menu ordering - APP_NAME Constant (e.g.,
APP_NAME = "workflow_genesis_internal"
): Internal identifier that MUST REMAIN STABLE
4. State Management via DictLikeDB
- State stored as JSON blobs in pipeline table
- Accessed via
pip.get_step_data()
andpip.set_step_data()
- All state changes are transparent and observable
5. Plugin Discovery System
๐ PLUGIN DISCOVERY SYSTEM
plugins/
โโโ 010_introduction.py โ
Registered as "introduction" (menu order: 1)
โโโ 020_profiles.py โ
Registered as "profiles" (menu order: 2)
โโโ hello_flow (Copy).py โ SKIPPED - Contains "()"
โโโ xx_experimental.py โ SKIPPED - "xx_" prefix
โโโ 200_workflow_genesis.py โ
Registered as "workflow_genesis" (menu order: 20)
๐ AUTO-REGISTRATION RULES:
โ
Numeric prefix โ Menu ordering + stripped for internal name
โ Parentheses "()" โ Development copies, skipped
โ "xx_" prefix โ Work-in-progress, skipped
๐ง Must have: landing() method + name attributes
๐ Auto dependency injection via __init__ signature
- Files in
plugins/
directory are auto-discovered - Numeric prefixes control menu ordering
- Classes must have
landing
method and name attributes - Automatic dependency injection based on
__init__
signature
Workflow Development Helper Scripts
Pipulate includes sophisticated helper scripts for workflow development:
create_workflow.py
Creates new workflows from templates:
python create_workflow.py workflow.py MyWorkflow my_workflow \
"My Workflow" "Welcome message" "Training prompt" \
--template trifecta --force
splice_workflow_step.py
Adds steps to existing workflows:
python splice_workflow_step.py workflow.py --position top
python splice_workflow_step.py workflow.py --position bottom
Template System
๐๏ธ WORKFLOW TEMPLATE SYSTEM
โโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโ
โ BLANK TEMPLATE โ โTRIFECTA TEMPLATEโ
โโโโโโโโโโโโโโโโโโโค โโโโโโโโโโโโโโโโโโโค
โ โโโโโโโโโโโโโโโ โ โ โโโโโโโโโโโโโโโ โ
โ โ Step 1 โ โ โ โ Step 1 โ โ
โ โ (Minimal) โ โ โ โ (Input) โ โ
โ โโโโโโโโโโโโโโโ โ VS โ โโโโโโโโฌโโโโโโโ โ
โ โ โ โ โ
โ Quick Start โ โ โผ โ
โ Single Purpose โ โ โโโโโโโโโโโโโโโ โ
โโโโโโโโโโโโโโโโโโโ โ โ Step 2 โ โ
โ โ (Process) โ โ
create_workflow.py โ โโโโโโโโฌโโโโโโโ โ
--template blank โ โ โ
โ โผ โ
โ โโโโโโโโโโโโโโโ โ
โ โ Step 3 โ โ
โ โ (Output) โ โ
โ โโโโโโโโโโโโโโโ โ
โ โ
โ Full Pattern โ
โ Complete Flow โ
โโโโโโโโโโโโโโโโโโโ
create_workflow.py
--template trifecta
blank
: Minimal workflow with one steptrifecta
: Three-step workflow pattern- Automatic method generation and insertion
Workflow Reconstruction System
The Revolutionary Alternative to OOP Inheritance: Atomic transplantation of workflow components using intelligent pattern matching and AST precision.
๐งฌ WORKFLOW RECONSTRUCTION: ATOMIC TRANSPLANTATION
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
OLD WORKFLOW WORKFLOW UPDATED WORKFLOW
(Atomic Source) RECONSTRUCTOR (Incremental Gen)
โโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโ
โ ๐งฌ Components: โ AST โ ๐ฏ Pattern โ AST โ โจ Generated: โ
โ โ โโโโบ โ Matching โ โโโโบ โ โ
โ โโโโโโโโโโโโโโโ โ โ โ โ โโโโโโโโโโโโโโโ โ
โ โstep_params* โ โ โ Bundle Type 1: โ โ โstep_params* โ โ โ
โ โstep_optim* โ โ โ Auto-Registered โ โ โstep_optim* โ โ โ
โ โparameter* โ โ โ Methods โ โ โparameter* โ โ โ
โ โโโโโโโโโโโโโโโ โ โ โ โ โโโโโโโโโโโโโโโ โ
โ โ โ Bundle Type 2: โ โ โ
โ โโโโโโโโโโโโโโโ โ โ Custom Routes โ โ โโโโโโโโโโโโโโโ โ
โ โ_process โ โ โ (_process, โ โ โ_process โ โ โ
โ โpreview โ โ โ preview) โ โ โpreview โ โ โ
โ โโโโโโโโโโโโโโโ โ โ โ โ โโโโโโโโโโโโโโโ โ
โโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโ
๐ COMPLETE LIFECYCLE: Test โ Validate โ Production โ Cleanup
--suffix 5 --target new_name --target same_name git status
โโโโโโโโโโ โโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโ โโโโโโโโโโ
param_buster5 advanced_params param_buster (in-place) (shows cruft)
(safe testing) (new workflow) (git history preserved) (clean up!)
๐ฏ WHY IT WORKS: Lightning in a Bottle
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ โจ Pattern Matching: No manual markers needed โ
โ ๐ง AST Precision: Syntactically perfect code generation โ
โ ๐ญ Inheritance Alternative: Compose without complex super() chains โ
โ ๐งช Safe Testing: Incremental validation without production risk โ
โ ๐ Git Continuity: In-place updates preserve development history โ
โ ๐งน Systematic Cleanup: Prevents file cruft accumulation โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
workflow_reconstructor.py --template botify_trifecta
--source parameter_buster
--suffix 5
The System That Eliminates Bootstrap Paradox:
- Atomic Sources: Battle-tested workflows become component libraries
- Pattern Matching: Intelligent detection via
_process
,preview
patterns - AST Transplantation: Surgical precision without syntax errors
- Complete Lifecycle: Development โ Testing โ Production โ Cleanup
๐ Quick Reference Card
Essential Commands
# Development workflow
cd ~/pipulate && nix develop # Start Pipulate
nix develop .#quiet # Start without auto-services
python server.py # Manual server start
git pull && nix develop # Update to latest
# Create new workflows
python helpers/workflow/create_workflow.py my_workflow.py MyClass my_internal_name
python helpers/workflow/splice_workflow_step.py my_workflow.py --position top
# Plugin naming conventions
010_my_plugin.py # Active plugin (menu order 1)
xx_my_plugin.py # Disabled during development
my_plugin (Copy).py # Ignored development copy
Key URLs & Ports
- Pipulate App:
http://localhost:5001
- JupyterLab:
http://localhost:8888
- Local AI Chat: Built into the Pipulate interface
- Logs:
tail -f logs/server.log
for debugging
Critical Patterns for AI Assistants
# Auto-key generation flow
if not user_input:
response = Response('')
response.headers['HX-Refresh'] = 'true'
return response
# Workflow initialization
return pip.run_all_cells(app_name, steps)
# State management
data = pip.get_step_data(step_id)
pip.set_step_data(step_id, updated_data)
File Structure Quick Reference
plugins/ # Your workflows (auto-discovered)
โโโ 010_introduction.py # Menu order 1
โโโ xx_draft.py # Disabled (xx_ prefix)
โโโ draft (Copy).py # Ignored (parentheses)
mcp_tools.py # AI assistant interface
common.py # Base classes for workflows
browser_automation/ # Selenium automation tools
logs/server.log # Debug everything here
data/data.db # SQLite application state
Common LLM Implementation Mistakes
๐จ LLM IMPLEMENTATION MISTAKE PREVENTION
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ COMMON PITFALLS โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ โ Missing HX-Refresh โ โ
if not user_input: โ
โ Response โ response.headers['HX- โ
โ โ Refresh'] = 'true' โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโผโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ โ Wrong Key Generation โ โ
pip.generate_pipeline_ โ
โ Method โ key(self) โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโผโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ โ Broken Chain Reaction โ โ
hx_trigger="load" โ โ
โ Pattern โ Automatic progression โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโผโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ โ APP_NAME Changes โ โ
NEVER modify after โ
โ (Data Orphaning) โ deployment โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
LLMs frequently make these errors:
- Missing HX-Refresh Response: Forgetting to return the refresh response for empty keys
- Incorrect Key Generation: Not using
pip.generate_pipeline_key(self)
properly - Missing Cursor Positioning: Forgetting the
_onfocus
attribute for user experience - Wrong Route Handling: Not understanding the difference between landing page and init routes
- State Inconsistency: Not properly handling the key generation and storage flow
- APP_NAME Changes: Modifying APP_NAME after deployment, orphaning existing data
- Chain Reaction Breaks: Not properly implementing the HTMX step progression pattern
Key Design Guidelines & Patterns
These "speedbumps" reinforce Pipulate's core philosophy:
- Local vs. Enterprise Mindset: Embrace local-first simplicity. Avoid patterns designed for distributed, multi-tenant systems.
- JSON State Management (Workflows): Keep workflow state in self-contained steps within a single JSON blob per run. Avoid complex state machines or external step tracking.
- Database (MiniDataAPI): Use the simple schema definition and access patterns provided. Avoid heavy ORMs.
- Workflow Pattern: Ensure workflows are linear and state is explicitly passed or saved at each step. Avoid complex async task chaining that obscures state.
- UI Rendering Pattern: Generate HTML directly from Python components via FastHTML. Avoid template engines.
- WebSocket Pattern: Use the dedicated
Chat
class for managing LLM interactions. Avoid raw WebSocket handling elsewhere. - Workflow Progression Pattern: Workflows use an explicit chain reaction pattern with
hx_trigger="load"
to manage step progression. This pattern must be preserved exactly as implemented. See the workflow documentation for details.
Internal Components
- Monitoring: A file system watchdog monitors code changes. Valid changes trigger an automatic, monitored server restart via Uvicorn, facilitating live development.
โโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโ
โ File System โ Changes โ AST Syntax โ Checks Code
โ Watchdog โ Detects โ Checker โ Validity
โโโโโโโโฌโโโโโโโ โโโโโโโโโฌโโโโโโโ
โ Valid Change โ
โผ โผ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโ
โ Uvicorn Server โโโโโ โ Reload โ Triggers Restart
โ (Handles HTTP, WS, SSE) โ โ Process โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโ
Pipeline State Inspector & MCP Tools
The system provides comprehensive debugging and state inspection capabilities through MCP tools and real-time monitoring:
๐ PIPELINE STATE INSPECTOR
โโโโ ๐ Discovering active workflows...
โโโโ โก Found 3 running processes
โโโโ ๐ฏ Step 2/5: Processing data transformations
โโโโ โจ Ready for next interaction!
๐ค MCP TOOL CALLS
โโโโ ๐ก Connecting to Botify API...
โโโโ ๐ Authentication successful
โโโโ ๐ Fetching schema (4,449 fields discovered!)
โโโโ ๐พ Caching results for lightning-fast access
The Deeper Philosophy: Why This Matters
Pipulate represents more than technology choices โ it's a statement about computing sovereignty and the future of human-AI collaboration.
The Bootstrap Paradox Solved
How do you "lift yourself up by your bootstraps"? The answer: you're not a closed system. Humans interact with reality through interfaces, just like software components. We are "human apps with reality APIs."
This pattern repeats at every scale โ from quantum fields to societies. The art lies in choosing the right granularity for these interfaces. Too abstract, and you lose control. Too granular, and you drown in complexity.
Every API is a Compromise
The most successful APIs are durable โ they become the "laws of physics" for their domain. In reality: the laws of physics, chemistry, biology. In tech: logic gates, operating systems, protocols like HTTP and HTML.
Pipulate chooses these durable APIs deliberately, creating a future-proof foundation for AI-assisted development that will outlast the current framework churn.
Writing for the Future
This documentation serves multiple audiences: humans seeking immediate productivity, and AI systems learning to collaborate with human creativity. The explicit, observable nature of Pipulate workflows creates perfect training data for the next generation of AI assistants.
Final Thoughts
Pipulate offers a unique approach to building local, AI-assisted applications, prioritizing simplicity, user control, and reproducibility over conventional scaling patterns. By understanding its core concepts and embracing its philosophy, developers can create powerful workflows, and end-users can benefit from AI guidance with complete flexibility in deployment options.
Remember the guiding principles:
- Keep it simple.
- Support local-first with cloud integration options.
- Embrace the unconventional while remaining practical.
- Choose durable foundations that work with any approach.
- Build for both human creativity and AI collaborationโlocal or cloud.
The Bottom Line: Pipulate doesn't reject the modern AI ecosystemโit provides a structured foundation that works with any AI service. Whether you're using Claude via API, ChatGPT for reasoning, or local models for privacy, Pipulate gives you the workflow framework to orchestrate them all effectively. It's not about choosing sides in the AI warsโit's about having the right tool for any job.
Developer's Notes
The Pipulate Workshop
The repository includes not only polished plugins but also experimental scripts and notebooks under development (e.g., in the root directory or marked with xx_
prefix in plugin directories). These represent ongoing work and exploration.
Plugin Development Conventions
Auto-Registration Behavior
- Numeric Prefixes: Files like
workflows/10_hello_flow.py
are registered ashello_flow
(number stripped for internal name, used for menu order). - Parentheses Skip: Files with
()
in the name (e.g.,hello_flow (Copy).py
) are skipped โ useful for temporary copies during development. xx_
Prefix Skip: Files prefixed withxx_
(e.g.,xx_experimental_flow.py
) are skipped โ useful for keeping unfinished work in the plugin directories without activating it.
Workflow for Creating New Plugins
- Copy: Copy a template to
my_flow (Copy).py
. - Modify: Develop your workflow. It won't auto-register yet.
- Test: Rename to
xx_my_flow.py
. The server should auto-reload. Test thoroughly. - Deploy: Rename to
##_my_flow.py
to assign menu order and activate.
Git History Considerations
Use git mv
for simple renames (like xx_
to numbered prefix) to preserve history. Document more complex renames in commit messages.
git mv workflows/xx_my_flow.py workflows/##_my_flow.py
git commit -m "Feat: Promote workflow xx_my_flow.py to ##_my_flow.py"
About This README: Single Source of Truth Documentation
This README serves as the upstream source of truth for all Pipulate documentation across GitHub, Pipulate.com, and the built-in app documentation. Changes made here automatically cascade to all other documentation surfaces.
The ASCII Art Synchronization System
๐ THE UPSTREAM TRUTH CASCADE
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
๐ Source Code Reality (The Ultimate Truth)
โ
โผ
๐ README.md (Single Source of Truth)
โโ ASCII Art Blocks (Visual Truth)
โโ HTML Comment Keys (Metadata)
โโ 80-Hyphen Pagination (Structure)
โ
โโโโโโโโโโโโโโโโผโโโโโโโโโโโโโโโโโโโ
โผ โผ โผ
๐ GitHub Page ๐ Pipulate.com ๐ง Built-in Docs
(Auto-display) (Jekyll Build) (Live Integration)
โ โ โ
โผ โผ โผ
๐ Screenshots ๐ฌ Demos ๐งช Tests
(Future) (Future) (Future)
How it works:
- ASCII Art Blocks: Visual diagrams are automatically extracted and distributed to other documentation files
- HTML Comment Keys: Headlines marked with
<!-- key: identifier -->
serve as reference anchors - 80-Hyphen Pagination: Section dividers enable automatic document structuring
- Automatic Synchronization: Running
python helpers/docs_sync/sync_ascii_art.py
updates all documentation
This creates "ASCII art peer pressure" โ when visual diagrams change, they compel surrounding text to be updated for consistency, ensuring documentation accuracy across the entire ecosystem.
Roadmap
Core & Workflow Enhancements:
- Dev, Test, and Prod database switching
- Saving source HTML and rendered DOM of any URL
- Botify data export CSV save (incorporating robust polling)
- Full web form field support (textarea, dropdown, checkboxes, radio buttons)
- Generic support for Anywidgets
- Utility for deleting garbage tables from plugin experimentation
AI / LLM Integration:
- LLM inspection of any local data object (RAG-style functionality)
- Various memory types for LLM context (vector embedding, graph, key/val-store)
- Enabling the local LLM to be an MCP Client
Automation & External Interaction:
- MCP Server for automated web browsing and similar tasks
Included PrismJS Highlighting
THEMES
- Okaidia ocodia 1.77KB
LANGUAGES
- CSS1.71KB
- Markup + HTML + XML + SVG + MathML + SSML + Atom + RSS4.64KB
- C-like0.83KB
- JavaScript6.18KB
- Bash + Shell + Shell zeitgeist87 8.96KB
- Diff uranusjr 1.33KB
- JSON + Web App Manifest CupOfTea696 0.58KB
- JSON5 RunDevelopment 0.52KB
- JSONP RunDevelopment 0.23KB
- Liquid cinhtau 2.56KB
- Lua Golmote 0.74KB
- Markdown Golmote 10.43KB
- Markup templating
- Mermaid RunDevelopment 3.03KB
- Nix Golmote 1.47KB
- Python multipetros 2.45KB
- Regex RunDevelopment 2.33KB
- YAML hason 3.11KB
PLUGINS
- Line Highlight11.66KB
- Line Numbers kuba-kubula 7.23KB
- Toolbar mAAdhaTTah 5.63KB
Contributing
Contributions are welcome! Please adhere to the project's core philosophy:
- Maintain Local-First Simplicity (No multi-tenant patterns, complex ORMs, heavy client-side state).
- Respect Server-Side State (Use DictLikeDB/JSON for workflows, MiniDataAPI for CRUD).
- Preserve the Workflow Pipeline Pattern (Keep steps linear, state explicit).
- Honor Integrated Features (Don't disrupt core LLM/Jupyter integration unless enhancing local goals).
License
This project is licensed under the MIT License. See the LICENSE file for details.
Resources
Background Articles: Mike Levin, AI SEO in NYC
Enhanced Documentation: Pipulate AI SEO Software
On GitHub: Pipulate on GitHub
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file pipulate-1.1.1.tar.gz
.
File metadata
- Download URL: pipulate-1.1.1.tar.gz
- Upload date:
- Size: 570.8 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.1.0 CPython/3.12.11
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 |
6cfcdef678b5c1a1017fdd301071285936aec3696309aa32055967b2c02a0804
|
|
MD5 |
246b9acfbb195434ac7929b0eb6130cb
|
|
BLAKE2b-256 |
5d351b92f28752d6b2cc54251deb1fcae0bd4c7fe00f44267e49e278e8a8a475
|
File details
Details for the file pipulate-1.1.1-py3-none-any.whl
.
File metadata
- Download URL: pipulate-1.1.1-py3-none-any.whl
- Upload date:
- Size: 579.0 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.1.0 CPython/3.12.11
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 |
2aebccc2c998f544fb0546ac8c068dbed7241ac04891a3a98aec9445f21acadb
|
|
MD5 |
34dd6168b60f378fe9c031331e8a0fc7
|
|
BLAKE2b-256 |
ccfd8d2b3c837e151bb9df72276bc9725ec79e9124b6e69e3bece337cbafebd9
|