MoonshotAI/kimi-cli
Kimi Cli
Kimi is a terminal-based AI agent that autonomously plans and executes software development tasks by reading, editing, and running code. It operates as an intelligent command-line agent that breaks down high-level goals into sequences of shell commands and code edits, carrying them out without manual step-by-step guidance. The agent can run in an interactive loop, switch to a shell mode for direct terminal command execution, and operate in non-interactive or one-shot modes suitable for scripting.
The project distinguishes itself through multiple integration and execution modes. It can run as an Agent Communication Protocol (ACP) server, allowing any ACP-supporting editor or IDE to invoke it, and offers a dedicated VS Code extension for seamless code editing within the editor. The agent supports plan-based autonomous execution, where it breaks down goals into sub-steps and executes them by reading, editing, and running code. It also provides a browser-based OAuth authentication flow for accessing user accounts and available models, and can connect to external tools and services through the Model Context Protocol with configurable timeouts.
The CLI supports extensive configuration and extensibility, including file-based settings loading from TOML or JSON files, agent personality selection, API provider configuration, model selection, and custom skills directories that the agent automatically discovers and loads at startup. It includes lifecycle hooks that run shell commands on agent events, background task management with configurable concurrency and timeouts, and session management features for saving, resuming, and exporting sessions. The agent also offers a web UI for remote interaction and trace visualization, and an AI-enhanced Zsh plugin that adds agent capabilities to the shell.
Features
- Agentic Terminal Environments - Runs an AI agent directly in the terminal to execute shell commands and automate development workflows.
- Agentic Code Editing - Breaks down high-level goals into shell commands and code edits for automated software workflows.
- Autonomous Task Execution - Breaks down a software task into sub-steps and carries them out autonomously by reading, editing, and running code.
- Agent Communication Protocol Servers - Runs as an Agent Communication Protocol server so any ACP-supporting editor or IDE can invoke it.
- Automated Code Editing Assistants - Opens source files, interprets their content, and applies changes such as inserting, deleting, or replacing lines of code.
- CLI Agent Management - Provides an intelligent command-line agent that plans and executes software development tasks autonomously.
- Shell Command Execution - Executes terminal commands directly from the agent's plan, including installing dependencies and running tests.
- Interactive Shell Loops - Runs an interactive agent loop that can switch to a shell mode for direct terminal command execution.
- Autonomous Planning Engines - Breaks down high-level goals into a sequence of sub-steps and executes them autonomously.
- Coding Automation Tools - Reads, edits, and runs code in a terminal environment to automate software development workflows.
- Browser-Based OAuth Flows - Logs into or out of a user account through a browser-based OAuth flow to access available models.
- AI Agent Sessions - Saves, resumes, and exports AI agent sessions for debugging, sharing, or continuing work.
- Model Context Protocol Integrations - Connects to external tools and services through a standardized protocol for tool calls with configurable timeouts.
- ACP-Compatible Servers - Implements the Agent Communication Protocol for integration with editors and IDEs.
- Execution Step Limits - Sets limits on execution steps, retries, and context compaction thresholds to manage agent runtime.
- Iteration Limits - Limits the number of steps per turn, retries per step, or iterations in a loop mode.
- MCP Server Management - Adds, authorizes, and connects to Model Context Protocol servers using sub-commands or a config file.
- Web Search & Fetch Configurations - Configures web search and fetch services to retrieve URLs and search the web.
- MCP Server Configurations - Loads Model Context Protocol configuration from files or JSON strings to connect external tools.
- Default Model Selection - Overrides the default language model used for generating responses in a session.
- Provider Configurations - Defines API provider connection details including base URL, API key, and custom headers for model access.
- ACP-Compatible Servers - Starts the agent as an ACP-compatible server so any ACP-supporting editor or IDE can use it.
- CLI Prompt Executions - Ships a one-shot mode that feeds a single prompt and exits, suitable for scripting.
- Non-Interactive CLI Modes - Outputs results as plain text or JSONL without an interactive interface, suitable for scripting.
- VS Code Extensions - Connects an AI agent to Visual Studio Code through an extension for seamless code editing.
- Working Directory Configuration - Defines the root directory for file operations and expands the workspace scope with additional directories.
- Configuration File Loading - Reads settings from TOML or JSON files or inline strings to customize CLI behavior.
- Browser-Based Authentication Strategies - Logs into user accounts through a browser-based OAuth flow to access available models and services.
- Directory-Based Plugin Discovery - Loads agent skills from configurable directories that the CLI automatically discovers and registers at startup.
- Read-Only Planning Sessions - Provides a plan mode that restricts the agent to read-only exploration for producing implementation plans.
- Coding Assistants - CLI interface for accessing advanced language models.
- Command Line Tools - Official CLI for an AI assistant focused on coding tasks and workflows.
