MindPalace: A Local-First AI Assistant Framework

MindPalace is an open-source AI assistant engineered to prioritize user autonomy and privacy. By granting full control over data and operations, it empowers individuals to customize their AI experience without external dependencies.

Steam-Friendly Packaging

Ready-to-ship runtime payloads (Ollama + Whisper) now live under runtime/, with a cross-platform bootstrapper located at cmd/bootstrapper. See packaging/steam/README.md for the full staging + SteamPipe upload flow, including how to prefetch Ollama releases, bundle models, and configure Steam launch options so clean installs work without manual dependency setup.

Key Features

• Local Execution: Runs entirely on your device using lightweight, unbiased language models, ensuring data privacy and independence from cloud services or corporate ecosystems.
• Event-Sourced Architecture: Built on a robust, efficient foundation that maintains a complete history of interactions for reliability and easy recovery.
• Modular Plugin System: Extends core functionality through agent-based plugins, such as task management or code assistance, allowing seamless integration of new capabilities.
• Multi-Modal Input: Supports voice and text-based interactions for intuitive configuration and control.
• Extensibility: Designed for adaptability, enabling users to tailor the system to their unique workflows and projects.

Philosophy

MindPalace shifts the paradigm from prescriptive AI tools to a flexible framework where the computer adapts to the user. It provides a streamlined, professional-grade toolset for building personalized digital solutions, ideal for developers, creators, and power users seeking a trustworthy AI companion.

Getting Started

1. Clone the Repository:

   git clone https://github.com/yourusername/mindpalace.git

2. Install Dependencies: Refer to requirements.txt for Python dependencies and go.mod for Go modules.
3. Build and Run: Use the provided Makefile or run directly via:

   go run cmd/mindpalace/main.go

3D Asset Pipeline

MindPalace now ships with a baked catalog of Godot-ready meshes. To rebuild or curate the library:

• Convert the Thingi10K archive into GLB assets and a manifest:

  go run cmd/modelbaker/main.go -tar Thingi10K-002.tar.gz

• Assets are written under world/assets/models, and the manifest lands in catalog.json. Godot consumes the res://assets/models/<id>.glb paths directly—no runtime extraction.
• Optional -metadata path/to/models.json lets you supply human-friendly titles and tags (JSON mapping model_id → { "title": "...", "tags": [...] }) which power search during orchestration and plugin execution.
• At runtime the orchestrator loads the manifest; plugins can request models by id and capture scale hints and bounding boxes from the catalog.

Contributing

We welcome contributions to enhance MindPalace. Please review our code of conduct, submit issues for bugs or features, and open pull requests for improvements.

License

MindPalace is licensed under the MIT License. See LICENSE for details.

For more information, explore the documentation or contact the development team.