Summary Codemini-CLI offers a "terminal-first AI workspace" designed for maximum compatibility and minimal overhead. It functions as a coding and task automation CLI, supporting any OpenAI-compatible or Anthropic API across Windows, macOS, and Linux. Its core benefit is providing powerful agentic assistance without wasting context budget, making AI tools more accessible and efficient in diverse terminal environments.

How It Works Codemini adopts a "restrained" architectural philosophy to optimize AI interaction. Key features include dual-model dispatch, routing simple tasks to a lightweight model and complex reasoning to a frontier model. Skills are lazy-loaded, meaning only invoked skills contribute to the prompt, drastically reducing token usage. Memory is curated through a "dream loop" and self-evolution via /reflect, converting successful workflows into reusable skills. Project context is managed via a compact index built with Tree-sitter AST parsing, replacing large file trees with symbol maps.

Quick Start & Requirements Install globally via npm: npm install -g codemini-cli. Configuration requires setting up an OpenAI-compatible or Anthropic API gateway URL and API key. Users must also specify their desired model name. Optional accelerators like fff-mcp and playwright can be installed for enhanced performance.

Highlighted Details

• Efficient Context Management: Achieves significantly lower system prompt token counts (~5,300 tokens) compared to alternatives by employing lazy-loaded skills, curated memory, and compact project indexing.
• Dual Interfaces: Offers both a rich Terminal TUI and a browser-based Web UI, sharing the same session state and engine.
• Extensible & Adaptive: Features a skills system with lazy loading, persistent memory that self-evolves into reusable skills via /reflect, and configurable "Souls" for agent tone.
• Controlled Execution: Includes planning mode, proportional approvals for destructive actions, and session checkpoints for state persistence.
• Broad Compatibility: Runs on Windows, macOS, and Linux, supporting various shells and AI providers.

Maintenance & Community The provided README does not detail specific community channels, active contributors, or roadmap information.

Licensing & Compatibility Licensed under the MIT license, Codemini is designed for broad compatibility, including commercial use, as it runs locally without SaaS, telemetry, or mandatory registration. It supports any OpenAI-compatible or Anthropic API.

Limitations & Caveats Functionality is dependent on the configuration and availability of external LLM APIs. Performance for JavaScript-heavy pages may benefit from optional Playwright installation. The "restrained" design prioritizes efficiency over feature bloat, which might mean fewer built-in tools than some alternatives.