GCWing/BitFun

What it solves

BitFun addresses the limitation of traditional AI chat interfaces by transforming tasks into dedicated "Mini Apps" with their own UIs. It also solves the problem of vendor lock-in and privacy in multi-device agent control by providing a self-hostable, zero-knowledge relay server, and tackles the high cost and latency of long-context agent tasks through optimized KV cache reuse and fast repository searching.

How it works

BitFun operates as a desktop AI agent with a Rust runtime that can drive the browser, terminal, filesystem, and other desktop applications. It uses a byte-stable prompt assembly to maintain a high KV cache hit rate and a specialized indexing tool called "flashgrep" to accelerate context retrieval in massive codebases. Users can extend the system through four tiers of customization, ranging from simple custom agent definitions to full source-level changes.

Who it’s for

It is designed for software engineers needing an agent for complex coding tasks (planning, debugging, and deep review) and office professionals who need to turn source materials into deliverables like reports, presentations, and meeting notes.

Highlights

• Agentic Mini Apps: Creates task-specific interfaces (charts, boards, forms) instead of relying solely on a chat box.
• Self-Hosted Control: Allows cross-device session sync and remote control via a user-deployed relay server for maximum privacy.
• High Efficiency: Achieves a 98.67% average KV cache hit rate and up to 36x faster searches in large repositories via flashgrep.
• Extensible Architecture: Supports MCP, Skills, and Codex-compatible Hooks for integrating external tools.