XOne

The local AI control plane for models, knowledge, agents, and production workflows.

Download Desktop · Build Status · Support

LOCAL-FIRST   MODEL-AGNOSTIC   MCP-READY   OPENAI-COMPATIBLE

XOne brings local inference, private knowledge retrieval, Model Context Protocol (MCP) tools, training, export, and OpenAI-compatible serving into one observable workspace. X1 is the compact mark; XOne is the product.

_{One workspace. Observable local intelligence.}

Technology Stack

Product Surface

Download XOne Desktop

The latest release provides installers for the supported platforms:

Verify downloads with the published SHA256SUMS.txt.

Installation

• macOS: Open the DMG and drag XOne.app to Applications. Alpha builds may require System Settings → Privacy & Security → Open Anyway.
• Windows: Run the x64 installer. Review the SmartScreen warning and verify the checksum before continuing.
• Debian / Ubuntu: sudo apt install ./XOne-Desktop-Linux-x64.deb
• Linux AppImage: chmod +x XOne-Desktop-Linux-x64.AppImage && ./XOne-Desktop-Linux-x64.AppImage

Quickstart

macOS, Linux, and WSL

git clone https://github.com/abhishekpandaOfficial/XOne.git
cd XOne
./install.sh --local
xone studio

Windows PowerShell

git clone https://github.com/abhishekpandaOfficial/XOne.git
cd XOne
.\install.ps1 --local
xone studio

The local service listens on 127.0.0.1:8888 by default. For a secure API-only session:

xone studio --api-only --secure -p 8888

Useful local endpoints:

• Health: http://127.0.0.1:8888/api/health
• OpenAPI: http://127.0.0.1:8888/openapi.json
• Swagger UI: http://127.0.0.1:8888/docs
• ReDoc: http://127.0.0.1:8888/redoc

Models

XOne supports a model workflow built around the task, hardware, and context window:

1. Search the model hub or import a compatible local checkpoint.
2. Start with a smaller chat or instruct model while validating prompts and data.
3. Use quantization and runtime selection appropriate to the host hardware.
4. Keep model identity, residency, loading state, and request activity observable.
5. Scale to a larger or multimodal model when quality and context requirements justify it.

Model files remain on the configured local cache unless the operator explicitly chooses another storage or provider boundary.

MCP and Agents

MCP lets an agent call focused tools such as documentation search, repositories, files, or data systems. XOne keeps this integration explicit:

xone start claude
xone start codex
xone start opencode

Start with read-only servers, review tool permissions, keep credentials in the local credential store, and enable only the servers required by a workspace. The XOne Docs MCP preset is available from the MCP controls inside the workspace.

RAG and Private Knowledge

RAG is a controlled pipeline, not a model setting:

1. Ingest: collect authoritative PDFs, DOCX files, CSVs, URLs, or project data.
2. Prepare: normalize content, preserve metadata, and define access scope.
3. Retrieve: select the smallest relevant context for the request.
4. Generate: answer with source-aware context and citations where available.
5. Evaluate: measure retrieval quality, answer quality, latency, and drift independently.

Keep sensitive sources local, separate tenant or project indexes, and avoid sending retrieved context to an external provider unless that boundary is deliberate.

Training and Export

XOne supports a practical local lifecycle:

• Prepare and validate datasets with explicit splits and schemas.
• Use LoRA for efficient adapter training and QLoRA for lower-memory workflows.
• Inspect training loss, evaluation behavior, checkpoints, and reproducibility metadata.
• Export adapters, merged checkpoints, or GGUF artifacts for the target runtime.
• Validate the exported artifact through the local chat and API paths before deployment.

API and Operations

XOne exposes native /api/* endpoints and OpenAI-compatible /v1/* endpoints. A minimal local request looks like this:

curl http://127.0.0.1:8888/v1/chat/completions \\
  -H 'Content-Type: application/json' \\
  -d '{"model":"local-model","messages":[{"role":"user","content":"Hello"}]}'

Recommended production controls:

• Keep the service on loopback unless network access is required.
• Use HTTPS, authentication, strict CORS, and host policy for remote access.
• Treat MCP tools and retrieved documents as separate security boundaries.
• Do not place secrets in VITE_* variables or public frontend configuration.
• Monitor health, request activity, model residency, latency, and resource usage.
• Do not deploy the model runtime as a short-lived serverless function.

Architecture

X1-Studio Desktop / XOne Web Portal
                |
          React + Vite UI
                |
       Local FastAPI control plane
       /api/*          /v1/*
                |
  Model runtimes · RAG · MCP · Training · Export

Development

# Frontend
cd studio/frontend
npm ci
npm run dev -- --port 5173 --strictPort

# Validation
npm run typecheck
npm run build
npm test

# Native desktop development
cd ..
npx --yes @tauri-apps/cli@2.10.1 dev

Run backend development from the repository root with:

./.venv/bin/xone studio --api-only -H 127.0.0.1 -p 8888

Pull requests should include a concise change description, focused tests, and screenshots for user-facing changes. Keep credentials, model weights, generated binaries, and machine-specific state out of commits.

Repository Structure

xone/
├── studio/frontend/       React web and desktop interface
├── studio/backend/        FastAPI control plane and runtime services
├── studio/src-tauri/      Tauri shell, icons, and native packaging
├── Python CLI package/     xone command implementation and compatibility layer
├── tests/                  Integration, security, and regression tests
├── scripts/                Build, packaging, and verification utilities
└── .github/workflows/      CI and desktop release automation

Releases

Desktop releases are built by the Release XOne Desktop Alpha workflow. Each release must publish all platform packages and SHA256SUMS.txt. The portal uses stable latest-release asset paths, so a new desktop release does not require a frontend version edit.

Before a production release, configure Apple Developer ID signing and notarization, a trusted Windows certificate, release monitoring, and a documented rollback plan.

License and Notices

See LICENSE, COPYING, and studio/LICENSE.AGPL-3.0 for the applicable license terms and notices. Review component-specific notices before redistributing builds.

XOne includes open-source components and technical foundations originally developed with Unsloth AI. See the repository notices for attribution and licensing details.