Koder

An experimental, universal AI coding assistant for the terminal. Built in Python, Koder works with OpenAI, Anthropic, Google, GitHub Copilot, and 100+ providers via LiteLLM.

🎯 Status: Alpha - This is a learning-focused project exploring AI agent development.

Features

• Universal AI Support - Works with ChatGPT/Gemini/Claude subscriptions and API keys via OpenAI, Anthropic, Google, GitHub Copilot, and 100+ providers
• Smart Context - Persistent sessions with SQLite storage and automatic token-aware compression
• Real-time Streaming - Rich terminal displays with live output
• Comprehensive Tools - File operations, search, shell, task delegation, todos, and skills
• MCP Integration - Extensible tool ecosystem via Model Context Protocol
• Zero Config - Automatic provider detection with sensible defaults

Installation

Using uv (Recommended)

uv tool install koder

Using pip

pip install koder

Quick Start

# 1. Set your API key (works with any provider)
export KODER_API_KEY="your-api-key"

# 2. Run Koder
koder

That's it! KODER_API_KEY works with any provider - no need to remember provider-specific variable names.

Basic Usage

# Interactive mode
koder

# Single prompt
koder "create a Python function to calculate fibonacci numbers"

# Named session (persists conversation)
koder -s my-project "help me implement a new feature"

# Use a different model
KODER_MODEL="claude-opus-4-20250514" koder "your prompt"

Configuration

Koder can be configured via (in priority order):

1. CLI arguments - Highest priority
2. Environment variables - Universal KODER_* vars override everything
3. Config file - ~/.koder/config.yaml

Environment Variables

| Variable | Description | Example | |----------|-------------|---------| | KODER_API_KEY | Universal API key (works with any provider) | sk-... | | KODER_MODEL | Model to use | gpt-4o, claude-opus-4-20250514 | | KODER_BASE_URL | Custom API endpoint | http://localhost:8080/v1 | | KODER_REASONING_EFFORT | For reasoning models | low, medium, high |

KODER_API_KEY and KODER_BASE_URL take priority over provider-specific variables (like OPENAI_API_KEY) and config file settings.

Providers

Use KODER_API_KEY for any provider, or provider-specific variables:

| Provider | Environment Variable | Model Example | |----------|---------------------|---------------| | Any | KODER_API_KEY | Works with all providers | | OpenAI | OPENAI_API_KEY | gpt-4o, gpt-4.1 | | Anthropic | ANTHROPIC_API_KEY | claude-opus-4-20250514 | | Google | GOOGLE_API_KEY | gemini/gemini-2.5-pro | | GitHub Copilot | (device auth) | github_copilot/claude-sonnet-4 | | Azure | AZURE_API_KEY | azure/gpt-4 |

See Configuration Guide for all 100+ providers.

OAuth Providers (Subscription-Based)

Use your existing subscriptions (Claude Max, ChatGPT Plus/Pro, Google Gemini) without API keys:

# Authenticate with a provider
koder auth login google      # Google/Gemini CLI (free with Google account)
koder auth login claude      # Claude Max subscription
koder auth login chatgpt     # ChatGPT Plus/Pro subscription
koder auth login antigravity # Antigravity (Gemini/Claude models)

# Check authentication status
koder auth list              # List configured providers
koder auth status            # Show detailed token status

# Revoke access
koder auth revoke google     # Remove stored tokens

OAuth Providers:

| OAuth Provider | Subscription | Description | |----------------|--------------|-------------| | google | Google Gemini | Access to Gemini models via Google account | | claude | Claude Pro/Max | Access to Claude models via subscription | | chatgpt | ChatGPT Plus/Pro | Access to GPT models via subscription | | antigravity | Antigravity | Access to Gemini/Claude models |

Available models are fetched from each provider's API after login and cached locally (1 day TTL). Use koder auth list to see all available models for your authenticated providers.

Usage after authentication:

# Use OAuth-authenticated models (OAuth provider prefix required)
KODER_MODEL="google/gemini-3-pro-preview" koder "your prompt"                # Google OAuth
KODER_MODEL="claude/claude-opus-4-5-20250514" koder "your prompt"            # Claude Max
KODER_MODEL="chatgpt/gpt-5.2" koder "your prompt"                            # ChatGPT OAuth
KODER_MODEL="antigravity/gemini-3-pro-high" koder "your prompt"              # Antigravity (Gemini 3.0)
KODER_MODEL="antigravity/claude-opus-4-5-thinking" koder "your prompt"       # Antigravity (Claude Opus 4.5)

OAuth tokens and cached models are stored in ~/.koder/tokens/ and automatically refreshed before expiry.

Config File Example

# ~/.koder/config.yaml

model:
  name: "gpt-4o"
  provider: "openai"
  reasoning_effort: null    # For reasoning models: low, medium, high

cli:
  session: null             # Default session name
  stream: true              # Enable streaming output

mcp_servers: []             # MCP server configurations

Commands

koder config show          # Show current config
koder config edit          # Edit config file
koder -s SESSION_NAME      # Use named session

MCP Servers

Model Context Protocol (MCP) servers extend Koder with additional tools.

CLI Commands

# Add servers
koder mcp add myserver "python -m my_mcp_server"
koder mcp add myserver "python -m server" -e API_KEY=xxx

# HTTP/SSE transport
koder mcp add webserver --transport http --url http://localhost:8000

# Manage servers
koder mcp list
koder mcp get myserver
koder mcp remove myserver

Config Example

# In ~/.koder/config.yaml
mcp_servers:
  # stdio transport (local command)
  - name: "filesystem"
    transport_type: "stdio"
    command: "python"
    args: ["-m", "mcp.server.filesystem"]
    env_vars:
      ROOT_PATH: "/home/user/projects"
    cache_tools_list: true
    allowed_tools:
      - "read_file"
      - "write_file"

  # HTTP transport (remote server)
  - name: "web-tools"
    transport_type: "http"
    url: "http://localhost:8000"
    headers:
      Authorization: "Bearer token123"

  # SSE transport
  - name: "streaming-server"
    transport_type: "sse"
    url: "http://localhost:9000/sse"

Skills

Skills provide specialized knowledge loaded on-demand, saving 90%+ tokens via progressive disclosure.

Directory Structure

Skills are loaded from (project skills take priority):

1. Project: .koder/skills/
2. User: ~/.koder/skills/

.koder/skills/
├── api-design/
│   └── SKILL.md
├── code-review/
│   ├── SKILL.md
│   └── checklist.md
└── testing/
    └── SKILL.md

Creating a Skill

Create a SKILL.md with YAML frontmatter:

---
name: api-design
description: Best practices for designing RESTful APIs
allowed_tools:
  - read_file
  - write_file
---

# API Design Guidelines

## RESTful Principles

Use nouns for resources, HTTP verbs for actions...

How Skills Work

1. Startup: Only skill names and descriptions are loaded (minimal tokens)
2. On-demand: Full content fetched when needed via get_skill(name)
3. Supplementary: Skills can reference additional files

Architecture

koder_agent/
├── agentic/        # Agent creation, hooks, and approval system
├── cli.py          # Main CLI entry point
├── config/         # Configuration management
├── core/           # Scheduler, context, streaming, security
├── mcp/            # Model Context Protocol integration
├── tools/          # Tool implementations
└── utils/          # Helpers and utilities

Core Flow

1. CLI (cli.py) parses arguments, initializes session
2. AgentScheduler (core/scheduler.py) orchestrates execution with streaming
3. Agent (agentic/agent.py) builds agent with tools, MCP servers, model settings
4. Tools (tools/engine.py) register tools, validate inputs, filter output
5. Context (core/context.py) persists conversations in SQLite

Data Storage

• Database: ~/.koder/koder.db (SQLite)
• Config: ~/.koder/config.yaml
• Skills: ~/.koder/skills/ or .koder/skills/

Development

Setup

git clone https://github.com/feiskyer/koder.git
cd koder
uv sync
uv run koder

Code Quality

uv run black .              # Format
uv run ruff check --fix     # Lint
uv run pytest               # Test

Security

• API Keys: Stored in environment variables, never in code
• Local Storage: Sessions stored in ~/.koder/
• No Telemetry: Only API requests to your chosen provider
• Shell Commands: Require explicit user confirmation

Contributing

Contributions are welcome!

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

See CONTRIBUTING.md for guidelines.

License

MIT License - see LICENSE for details.

---

<p align="center"> _{Built with Python and curiosity} </p>