Hi HN,

We’re building BitFun, an open-source Agentic Development Environment (ADE).

GitHub: https://github.com/GCWing/BitFun

Docs: https://deepwiki.com/GCWing/BitFun

BitFun explores a different way of working with AI during development. Instead of treating AI as a chatbox attached to an editor, it treats AI as an agent that can collaborate with you across the whole development workflow.

The system is built around an Agent architecture. Each user has an assistant that can evolve over time with memory and customizable behavior. On top of this foundation, BitFun currently includes:

Code Agent – an AI coding assistant

Cowork Agent – an AI assistant for knowledge work

Custom Agents – domain-specific agents defined via Markdown

For software development, the Code Agent provides several working modes:

Agentic Mode – day-to-day coding where the agent can read code, edit files, run commands, and verify results autonomously.

Plan Mode – for large tasks. The agent proposes a plan first, then executes after alignment.

Debug Mode – for tricky bugs. It instruments execution, traces code paths, compares behaviors, and helps identify root causes.

Review Mode – performs code reviews based on repository conventions and project rules.

BitFun is also designed to be extensible:

MCP protocol support for connecting external tools and resources

Skills: Markdown/script-based capability packages that teach the agent new tasks

Agent customization via Markdown definitions

Rules to inject repository conventions or behavior constraints (can import configs from tools like Cursor)

For model infrastructure, BitFun supports both local models and cloud APIs, so teams can switch providers easily depending on cost, performance, or privacy needs.

By default, code is not sent to remote services automatically. Users control which models run and where, and local models can run with full privacy.

Right now the desktop app (built with Rust + TypeScript / Tauri) supports macOS and Windows. We’re also exploring CLI, server mode, and integrations with platforms like Telegram or Discord.

This project started as an open-source exploration of human–AI collaborative development, and a large portion of the codebase was built through AI-assisted coding (“vibe coding”).

We’d love feedback from the community. Issues, ideas, and contributions are all welcome.