Getting started

Kodelet is an agent for software engineering and production operations. Use it to understand a codebase, implement a change, investigate a failure, or work through a longer task in your terminal, browser, or editor.

Start with a repository you trust and a small, verifiable task. Kodelet can edit files and execute commands, so review its changes as you would a teammate’s.

Installation

Kodelet supports macOS and Linux on amd64 and arm64. Choose one installation method.

Homebrew

brew tap jingkaihe/kodelet
brew install kodelet

Install script

curl -sSL https://raw.githubusercontent.com/jingkaihe/kodelet/main/install.sh | bash

The installer uses Homebrew on macOS and .deb or .rpm packages on Linux by default. To request a standalone binary instead:

curl -sSL https://raw.githubusercontent.com/jingkaihe/kodelet/main/install.sh | bash -s -- --binary

Check your installation:

kodelet version

Providers and authentication

For a fresh installation using the default OpenAI provider, set your API key before starting Kodelet:

export OPENAI_API_KEY="your-api-key"
export KODELET_WEAK_MODEL="gpt-5.6-luna"

Replace the example key with your own. This setup explicitly selects gpt-5.6-luna as the weak model. For Anthropic, subscription sign-in, or an OpenAI-compatible endpoint, follow provider configuration first. Choose models available to your account; see the model configuration examples.

Provider credentials belong to the daemon, not a remote client. If the local daemon is already running, finish active work and run kodelet server restart from the shell with the updated environment. Exporting a key in another terminal does not update an existing daemon.

Quick start

In your terminal

From your project’s directory, open the built-in terminal UI:

kodelet chat

Start with a prompt such as:

Explain this repository’s structure and how to run its tests. Do not edit files.

Then ask for a focused change and review the diff and test results. You can keep the conversation going without launching a new command for each message.

In your browser

Start the managed local daemon and open its Web UI:

kodelet server start
kodelet server url --open

Choose the built-in runner and your project directory when creating a conversation. The browser and terminal UI use the same local daemon and saved history. For foreground operation, kodelet serve is an alternative; see the Web UI guide.

Terminal chat automatically starts the daemon if needed. It remains running when you exit either UI, and conversations are saved automatically. Disconnecting and canceling are different actions.

For better project context, ask Kodelet to create an AGENTS.md and review its instructions. The customization guide also covers the built-in initialization recipe.

Usage modes

Configuration model

The daemon owns models, credentials, and conversation history. The runner owns workspace files and tools. The client selects a server and presents the interaction. Repository configuration cannot redirect clients, set daemon credentials, or widen trusted host permissions. See configuration and profiles before sharing a server or adding a remote runner.

Extensibility

Use AGENTS.md for project instructions, recipes for repeatable prompts, skills for model-invoked expertise, and extensions for tools and lifecycle behavior. Plugins package these capabilities for reuse. The customization guide walks through each option, including the TypeScript SDK.

Upstream docs

• Full user manual
• Sample configuration
• Source and contribution instructions

Kodelet is MIT-licensed.