Plugin (tunaLlama) that offloads heavy code generation from Claude Code and Codex CLI to a local LLM

A delegation tool designed to solve the token consumption and subscription limit problems that occur when using Claude Code or Codex CLI. It maximizes efficiency by separating model roles according to the nature of the task.

Role Division Structure

• Architect (Claude/Codex): Handles short input/output stages requiring high-level judgment, such as requirement decomposition, spec writing, and result verification.
• Developer (Local LLM): Handles long output stages with heavy token consumption, such as code generation, self-review, and revisions (using Ollama, LM Studio, etc.).

Key Features and Characteristics

• MCP (Model Context Protocol) based: Provides 13 tools through an MCP server, and the agent automatically calls tools as needed.
• Enhanced Korean Support: Uses the Kiwi morphological analyzer to improve the quality of Korean search and indexing for text without spacing.
• Efficient Context Management: All calls are logged in SQLite, and by breaking work units into small pieces and passing them to the local LLM, it reduces the context load on the upper-level model.
• Easy Installation: Asking the agent to run the installation guide automatically completes everything from dependency installation to plugin registration.

It is suitable for Claude Code Pro/Max users, developers who need to manage OpenAI API quotas, and users with a local LLM environment such as Ollama.