Native macOS app for running multiple Claude Code sessions in parallel. Tabbed interface with 7 themes, agent mode, persistent memory, usage tracking, and Wren compression built in.

Born from a simple problem: Claude Code is powerful but single-threaded. Complex projects need multiple agents working simultaneously -- one researching, one coding, one testing. Flock gives each agent its own tab with isolated context while sharing a unified interface.

FEATURES

• Tabbed multiplexing for parallel Claude Code sessions
• 7 built-in themes with pixel-accurate rendering
• Agent mode for autonomous multi-step workflows
• Persistent memory across sessions
• Real-time usage tracking and cost monitoring
• Wren compression integration for token savings
• Signed and notarized for macOS distribution

INSTALL

Download the latest .pkg or .zip from GitHub Releases. Signed and notarized for macOS.

LoRA fine-tuned 1.5B model for prompt compression. Achieves 60-80% token savings with minimal semantic loss. Runs locally on Apple Silicon via MLX with zero API costs.

Wren compresses at the input level rather than the model level. Instead of quantizing or pruning the model itself, it rewrites prompts and tool outputs to be more token-efficient while preserving the semantic content that drives model behavior. This means it works with any downstream model.

CAPABILITIES

• 60-80% token reduction on typical prompts
• Preserves semantic meaning and instruction signals
• Local inference via MLX on Apple Silicon
• MCP server for Claude Code integration
• Integrated into Flock for automatic compression
• Works with any downstream model or API

USAGE

# As MCP server in Claude Code
# Add to your claude config:
{
  "mcpServers": {
    "wren": {
      "command": "python3",
      "args": ["-m", "wren.mcp"]
    }
  }
}