Kimi K2 is a strong, low-cost coding model. agentty makes it a first-class terminal coding agent — you sign in with your existing Kimi plan over OAuth (no API key) and get the whole agent: native tool-calling, file edits, shell, retrieval, and streaming, from a single static binary with a millisecond cold start.
1. Install agentty
curl -fsSL https://agentty.org/install.sh | shOne static binary — no Node, no Python, no Electron.
2. Sign in with Kimi
agentty loginPick Kimi at the menu. agentty runs Kimi Code's OAuth device flow: it prints a one-time code and a URL. Approve in your browser and it switches the moment you're authorized. On SSH? Press c to copy the code to your local clipboard, paste it in any browser, and approve. The token is stored encrypted and refreshed automatically.
3. Code with Kimi K2
agentty --provider kimi--provider kimi persists, so after the first run you just type agentty. Switch provider live with ^P and model with ^/ — both have a search filter, so type kimi to jump straight to it.
Why Kimi K2 + agentty
- No API key — use the Kimi plan you already pay for, via OAuth sign-in.
- Native tool-calling — read/edit files, ripgrep search, shell, apply-patch, and subagents all run through standard OpenAI-style function calls.
- Streaming output and per-turn cost/token accounting.
- No lock-in — switch to DeepSeek, Claude, GPT, or Gemini live, no restart.
See the providers docs for details, or the deep-dive: Kimi K2 and DeepSeek in agentty.