# Koder

An experimental, open-source terminal AI coding assistant written in Python that combines a streaming TUI, persistent local sessions, scheduled goal-based runs, MCP integrations and sandbox-aware permissions.

Koder lets a developer bring their own model across OpenAI, Anthropic, Google, GitHub Copilot, Azure, OAuth-backed subscriptions, OpenRouter or more than 100 LiteLLM providers, worked from the terminal through slash commands, file mentions, shell mode, history search and optional voice dictation.

Session goals can carry token budgets and be paused, resumed, or scheduled as recurring local, cron-backed loops. Koder extends with project and user skills, plugins, MCP servers and background subagents that run local teams while the main session stays responsible for integrating their work, and stores all of its sessions, transcripts, memories, tasks and settings locally under its own paths, with model requests still going to whichever provider is configured. It installs with uv, and its status is alpha, described as a learning-focused project for exploring agentic coding systems.

Website: https://github.com/feiskyer/koder