About
Kodelet is a lightweight agentic SWE agent that runs in your terminal. It handles software engineering work, production operations tasks, and repository automation from a local CLI.
Key Features
One-shot and interactive: Use kodelet run for fast task execution or connect an
ACP client such as toad for longer interactive sessions.
Chat-based web UI: Run kodelet serve to launch a browser-based chat interface for
using Kodelet locally.
Git-aware workflows: Work with repositories directly from the CLI, including commit and pull request assistance.
Context-aware execution: Kodelet understands repositories, persists conversations, can follow up on previous runs, and supports headless JSON streaming for programmatic use.
Extensible: Add custom tools, agentic skills, lifecycle hooks, project context files, and plugin-based capabilities without changing the core binary.
Vision-ready: Analyze screenshots, diagrams, and mockups with image input support on vision-enabled models.
Principles
Kodelet is built around bring-your-own-key usage rather than subscription-gated limits. The focus is on making strong frontier models useful from the command line, with local project context and tool access.
Current documentation covers support for Anthropic, OpenAI-compatible models, and Google Gemini, with Claude vision support available today.
Common Commands
Installation
Package-based installation is the default path: Homebrew on macOS and .deb or .rpm
packages on Linux.
Documentation
The website now mirrors the current Kodelet docs with coverage for installation, usage modes, provider setup, automation workflows, configuration, skills, hooks, and custom tools.
Browse the documentation or jump to the upstream manual.
Resources
Source Code │ Documentation │ Blog │ Issues