Y̊pipe - Easy Local AI Client & MCP Orchestration Engine

Democratizing Enterprise-Grade Local AI.

<p align="center"> <img src="assets/logo-animation.svg" alt="Ypipe Logo Animation" width="200"> </p>

Y̊pipe is an easy-to-use local AI client and MCP orchestration engine—a secure, local desktop application that makes offline AI practical. Run optimized models (CPU/GPU) on your own hardware with absolute data sovereignty and zero cloud dependencies. Chain specialized models via Model Context Protocol (MCP) and bind them to on-premise systems to execute secure, autonomous workflows.

Note: Y̊pipe is the successor to Data-Philter. The legacy Data-Philter documentation and codebase are archived in archive/data-philter/README.md.

The key goals of this project are: Easy-to-use local AI, Data Sovereignty, Multi-Agent Systems, and Java-native stability - all on consumer enterprise hardware.

In the 2026 AI landscape, Ypipe acts as the Industrial Assembly Line for local AI.

---

🖥️ Interface Preview

Ypipe runs entirely on your local machine, bringing a visual agentic client, workflow builder, local model manager, and one-click integrations together. You can explore the full-size interactive interface at ypipe.com.

💬 Private Agentic Chat

Interact directly with your offline LLMs using natural language. Query local databases, fetch system information, and trigger automation scripts securely without transmitting data over the internet.

• Local LLM Chat: Talk directly to offline models running on your CPU or GPU.
• Agentic Gearbox Suggestion: Automatically recommends a fitting LLM model based on your system's CPU, GPU, and RAM capacity.
• Sovereign Data Execution: Run queries against SQLite or Apache Druid with zero external network requests.
• System Automation: Bind chat commands to trigger local scripts, legacy systems, and files safely.
• Zero Data Leaks: Your prompt, context, and responses never leave your physical machine.

🔀 SmartPipes Builder & Workflows

Design, test, and run visual DAG agentic pipelines. Chain offline LLMs, MCP tool calls, Javascript logic, and multi-engine processing.

• Visual DAG Flow Builder: Drag-and-drop workflow designer to assemble multi-step pipeline architectures.
• WebForm & REST API Publishing: Instantly make designed flows accessible as interactive webforms or headless REST endpoints to easily run and share workflows.
• Multi-Engine Pipeline Steps: Chain local GGUF models, MCP server calls, custom Javascript transformers, and media synthesis.
• Real-time Step Test & Cascading: Test individual pipeline steps with live input payload cascading without executing the full flow.

⚙️ Engine Foundry & Runtimes

Manage local runtimes and models with a built-in downloader, automatically assessing hardware to optimize CPU/GPU execution speed.

• Hardware-Optimized Config: Assess system memory to automatically enable Apple Silicon (Metal) or CPU/Vulkan acceleration.
• Model Downloader: Direct search and one-click import from Hugging Face for optimized GGUF formats.
• Versatile Model Support: Run specialized small models (800M parameters) up to larger reasoning models (31B+).
• Self-Contained Inference: No external inference engines (like Ollama) need to be installed. Everything is fully built-in and runs out-of-the-box.

🔌 MCP Couplings (Integrations)

Connect your offline AI agent to external tools, database servers, web browsers, and file managers instantly.

• Blueprints & McpIntegrations: Clear separation between static connection templates (Blueprints) and active running server instances (McpIntegrations).
• Dedicated Management UI: An intuitive, dedicated Couplings dashboard to manage, register, and monitor all active MCP servers and blueprints.
• Fine-Grained Tool Control: Full control over exposed tools, allowing you to disable specific tools, modify tool names, and rewrite descriptions.
• Zero-Dependency Runtimes: Run local servers using stdio, npx, uvx, or JBang directly on ypipe's embedded Java foundation.

---

🌊 Quick Start (JBang)

Run Ypipe instantly without installation using JBang:

jbang ypipe@iunera/ypipe

---

⛲ Installation

Grab the latest binaries for your OS at ypipe.com or download them directly from the GitHub Releases:

• Windows: Installer (.msi) or AppImage (.zip)
• macOS: Apple Silicon (.dmg)
• Linux: Ubuntu (.deb), RedHat (.rpm), or Tarball (.tar.gz)
• Universal: Executable JAR for any Java-enabled environment.

---

<img src="assets/watermill.svg" width="32" style="filter: invert(100%); vertical-align: middle;"> Why Ypipe?

Most AI tools are built for "Chat." Ypipe is built for Data Workflows. Ypipe targets connections to your system, your local secrets, and running on small devices.

---

<img src="assets/aqueduct.svg" width="32" style="filter: invert(100%); vertical-align: middle;"> Key Differentiators:

• Our goal is simple AI: We focus on ensuring that you do not need a doctorate to get a model and data docking running. We believe local AI needs to be accessible to everyone.
• The Intelligence Switchboard: Don't waste a 70B model on a 1B task. Ypipe lets you easily select models for sub-tasks. Imagine simple jobs utilizing tiny models for OCR or classification while reserving heavy compute for final synthesis. What works on your computer also runs easily on a server - this is how one gets AI into productive use.
• Built for the Professional Environment: Ypipe uses the Model Context Protocol (MCP) and standardized APIs to bridge local intelligence with your existing tools. Whether it's a local SQL database or a custom enterprise API, Ypipe provides the "dock" for your data.
• Java-Native Stability: No Python dependency hell. Ypipe is built on Java, fitting seamlessly into existing enterprise server architectures, high-security environments, and professional DevOps pipelines. The dependencies that are required for the outside world: Do not worry, Ypipe manages it.
• Sovereign Geopatriation: 100% of the data stays on your hardware. No cloud-routing, no data leaks, and zero external API costs. You own the infrastructure and the intelligence it produces.

---

<img src="assets/water-pump.svg" width="32" style="filter: invert(100%); vertical-align: middle;"> Features

• Autonomous Agent Chains: Orchestrate multiple local models to perform complex, multi-step tasks without human intervention.
• Semantic Continuity: Preserves context throughout the pipeline, ensuring that specialized agents have the "full picture" of the workflow.
• Cross-Platform UI & Headless: Run it with a full GUI on Windows/macOS/Linux or deploy it as a headless background service for automated enterprise logging and plausibility checks.
• OpenAI Compatibility: Drop-in replacement for OpenAI-based tools—just redirect the base URL to your local Ypipe instance.

---

<img src="assets/fire-hydrant.svg" width="32" style="filter: invert(100%); vertical-align: middle;"> Choosing Your AI Model

Ypipe removes the guesswork from local AI. You don't need to be a data scientist to select the right intelligence; Ypipe acts as a Model Switchboard, ensuring you use the most efficient engine for every specific task in your pipeline.

Ypipe features a built-in Gearbox that automatically assesses your local hardware (CPU, GPU, and RAM) and suggests appropriate models. Further, it auto-selects the best runtime for your hardware to run the models locally—ensuring that you do not need to deal with the complexity of local AI configuration.

---

<img src="assets/tap.svg" width="32" style="filter: invert(100%); vertical-align: middle;"> One-Click Connectivity

Ypipe is built to be the universal docking station for the Model Context Protocol (MCP). We have eliminated the "dependency hell" of manual installations.

• Zero-Dependency Setup: One-click installation of MCP servers. Ypipe handles the environment, runtimes, and connections—no additional software or Python dependencies are required on your host machine.
• Built-in Specialized Servers: Ypipe ships with a curated set of pre-defined MCP servers (including our own Druid MCP Server) for immediate enterprise utility. For community MCP servers, all MCP ser