Late CLI by Mlhher is a Go-based AI agent skill for Model Context Protocol (MCP) platforms. It provides a minimal, zero-config interface for connecting to various AI coding assistants and both local and cloud language models. Distributed as a single, statically compiled binary, it offers diverse installation methods across Linux, macOS, and Windows WSL. This tool aims to enhance developer throughput by orchestrating AI capabilities directly within the terminal, supporting complex agentic workflows for software development.

What does Late CLI connect to?

Late CLI functions as an AI agent skill connecting to AI coding assistants like Claude Code, Codex CLI, and ChatGPT. It integrates with various cloud providers including DeepSeek, Claude, GPT, Kimi, GLM, and OpenRouter. It also targets local models, specifically llama.cpp on port 8080 by default. The program supports Model Context Protocol (MCP) integration, allowing native mapping of external MCP servers via standard I/O for versatile model interaction.

Core features and tools

Within its terminal UI, Late CLI exposes several callable tools. These include /new to start a new session, /rewind for visual history, and /compose to draft long instructions externally. Users can select models with /model, view Git logs using /git, and see keybindings via /keybindings. Exiting is done with /exit. It also offers Context-Aware Search respecting .gitignore and .llmignore files, and File Attachments via "Ctrl+O".

Who is it for?

Late CLI targets "Builders wanting 10x throughput" and developers aiming to make local LLMs practical for agentic software development. It supports integrating powerful reasoning models with faster, cheaper local models for execution. This tool is for those who need to orchestrate complex code changes and plan implementations with AI assistance directly within their development environment. Skip this if your development workflow does not involve AI coding agents or local LLMs.

How to get started

For cloud model providers, the authentication model involves setting API keys via environment variables. For instance, OPENAI_API_KEY and OPENAI_BASE_URL configure communication via HTTP/HTTPS APIs.

Use cases

• Code Refactoring/Modification: Users can instruct the agent to add input validation or refactor database packages.
• Planning and Execution of Code Changes: Late CLI reads the codebase, generates an implementation plan, and seeks user approval before executing changes.
• Hybrid Model Routing: Architect plans with reasoning models (e.g., GPT 5.6) and spawn subagents to execute implementations using local models (e.g., Gemma 4).

Late CLI by Mlhher provides a terminal-based AI agent for developers seeking to integrate diverse AI coding assistants and language models into their workflow. It supports both local and cloud models with a zero-config approach for local setups and environment variable authentication for cloud services. The tool's capabilities extend to complex code changes, planning, and hybrid model use. Try starting a new session with "/new" to explore its interactive terminal UI.