Vibeathon Entry: This project is a submission for the BridgeMind Vibeathon (Feb 1-14, 2026)
BloxyCode is an enhanced fork of OpenCode featuring Bloxy - an autonomous task execution system that runs until your tasks are done.
An AI-powered coding CLI tool that brings the power of Claude and other LLMs directly to your terminal. BloxyCode enables autonomous code generation, intelligent file operations, and seamless AI-assisted development workflows.
BloxyCode's flagship feature - execute complex coding tasks autonomously from Product Requirements Documents (PRDs):
- Sequential Task Processing: Parse PRD markdown files and execute tasks one by one
- Automatic Testing: Run tests after each task (when configured)
- Auto-commit: Optional automatic git commits with descriptive messages
- Retry Logic: Configurable retry attempts for failed tasks
- Task State Management: Track pending, in-progress, completed, and failed tasks
- File Operations: Read, write, edit files with intelligent context awareness
- Code Search: Fast glob patterns and grep-based content search with ripgrep
- Bash Integration: Execute shell commands with proper error handling
- Web Capabilities: Web search and content fetching
- Task Management: Built-in todo list tracking for complex workflows
Create and manage specialized AI agents for different tasks:
- Primary Agents: Main agents for direct interaction
- Subagents: Specialized agents for delegated tasks
- Custom Tools: Configure which tools each agent can access
- Agent Generation: LLM-powered agent creation from descriptions
- Model Provider Support: Anthropic, OpenAI, Google, Azure, AWS Bedrock, and 15+ other providers
- MCP (Model Context Protocol): Standard protocol for AI-tool integration
- Custom Commands: Define project-specific slash commands
- Skill System: Reusable best practices and workflows
Access Claude and Gemini models through your Google account — no API keys required:
- Claude Opus 4.5 & Sonnet 4.5 with extended thinking support
- Gemini 3 Pro/Flash with configurable thinking levels
- Multi-account support with automatic rate limit rotation
- Free tier access using your Google Cloud quota
- See Antigravity Documentation for setup
- CLI Chat: Direct command-line interaction with AI
- Server Mode: Run as a persistent server with HTTP API
- TUI (Terminal UI): Rich terminal interface built with SolidJS
- Session Management: Continue conversations, attach to running servers
- Permission System: Fine-grained control over tool usage
- Pattern Matching: Allow/deny based on file patterns
- Interactive Prompts: Ask for permission when needed
# Install via npm
npm install -g @bloxystudios/bloxycode
# Or install via bun
bun install -g @bloxystudios/bloxycode# Clone the repository
git clone https://github.com/bloxystudios/bloxycode.git
cd bloxycode
# Install dependencies
bun install
# Run in development mode
bun run dev# Simple prompt
bloxycode run "create a new React component"
# Continue previous session
bloxycode run -c "add tests for the component"
# Use specific model
bloxycode run -m anthropic/claude-sonnet-4 "refactor the authentication logic"
# Attach files to context
bloxycode run -f src/app.ts "review this code for bugs"Create a PRD.md file:
# My Feature PRD
## Tasks
- [ ] Create user authentication module
- [ ] Add login and signup endpoints
- [ ] Write unit tests for auth moduleThen execute:
bloxycode run "/bloxy PRD.md"BloxyCode will autonomously:
- Read and parse the PRD
- Execute each task sequentially
- Run tests (if configured)
- Commit changes (if auto-commit enabled)
- Report progress and results
# Create a new custom agent
bloxycode agent create
# List available agents
bloxycode agent list
# Use a specific agent
bloxycode run --agent code-reviewer "review my changes"# Start server on specific port
bloxycode serve --port 4096
# Attach to running server
bloxycode run --attach http://localhost:4096 "your prompt"BloxyCode can be configured via:
~/.config/bloxycode/config.yml(global).bloxycode/config.yml(project-specific)
Example configuration:
# Default model
model: anthropic/claude-sonnet-4
# Share sessions automatically
share: auto
# Custom agents directory
agents: .bloxycode/agent
# Bloxy mode settings
bloxy:
autoCommit: true
testCommand: "bun test"
lintCommand: "bun run typecheck"
maxRetries: 3| Command | Description |
|---|---|
run [message] |
Start or continue a conversation |
serve |
Start the BloxyCode server |
agent create |
Create a new custom agent |
agent list |
List all available agents |
auth |
Authenticate with providers |
models |
List available models |
export |
Export session history |
import |
Import session history |
stats |
View usage statistics |
BloxyCode provides these tools to AI agents:
- bash: Execute shell commands
- read: Read file contents
- write: Create/overwrite files
- edit: Make precise edits to files
- glob: Find files by pattern
- grep: Search file contents
- webfetch: Fetch web content
- task: Delegate to subagents
- todowrite/todoread: Manage task lists
- bloxy_control: Control autonomous execution (Bloxy mode)
# Install dependencies
bun install
# Run in development mode
bun run dev
# Type check
bun run typecheck
# Run tests
bun test
# Run specific test
bun test test/tool/tool.test.ts
# Build
bun run buildsrc/
├── cli/ # CLI commands and interface
├── tool/ # AI tool implementations
├── session/ # Session management
├── agent/ # Agent system
├── bloxy/ # Autonomous mode implementation
├── provider/ # LLM provider integrations
├── server/ # HTTP server
└── util/ # Utilities
BloxyCode uses a modern TypeScript architecture:
- Runtime: Bun with TypeScript ESM modules
- Validation: Zod schemas for all inputs
- State: Namespace-based organization
- UI: SolidJS + OpenTUI for terminal interfaces
- Communication: MCP (Model Context Protocol) compatible
Contributions are welcome! Please see our contributing guidelines for more details.
MIT
- Built-in Antigravity models now work out of the box - Claude Opus 4.5, Claude Sonnet 4.5, and Gemini 3 Pro/Flash models available under Google provider without user configuration
- Fixed publish workflow - republish with all packages
- Added 14 new Bloxy slash commands for configuration, task sources, Git/PR integration, and utilities
- Added fallback AI models for release notes generation
- Bloxy Mode: Fixed PRD file checkbox not updating when tasks complete, fixed missing task context injection, and fixed autonomous task continuation (republish)
- Bloxy Mode: Fixed PRD file checkbox not updating when tasks complete, fixed missing task context injection, and fixed autonomous task continuation
- fix(bloxy): use directory fallback for non-git projects
- fix(config): add rules and capabilities to schema
- fix(config): add rules and capabilities to config schema
- fix(branding): final adjustments for upgrade command and docker paths
- fix: github organization name in publish scripts
- fix: docker image build paths
- feat: bloxy init, config, and github integration commands
- chore: release v1.0.5
- fix(publish): correct archive generation for scoped packages
- fix(publish): automate release process and fix authentication
- Fixed: npm global installation failing to find platform-specific binary
- Added
optionalDependenciesto package.json for all platform packages (darwin, linux, windows) - This allows npm to automatically install the correct binary for the user's platform
- Added
- Initial release of BloxyCode with Bloxy autonomous task execution
- Multi-provider AI support (Claude, OpenAI, Gemini, and 15+ more)
- MCP (Model Context Protocol) compatibility
- TUI interface with SolidJS
- Session management and agent system
- Documentation (coming soon)
- GitHub
- Discord Community (coming soon)
Built with ❤️ by Bloxy Studios