DeepSeek is one of the cheapest frontier-class coding models available. agentty makes it a first-class terminal coding agent: set your API key, point one flag at it, and get native tool-calling, file edits, shell, retrieval, and live-streamed reasoning — 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. Add your DeepSeek API key
export DEEPSEEK_API_KEY=sk-deepseek-YOURKEYagentty knows DeepSeek's endpoint natively — you don't configure a URL. It reads DEEPSEEK_API_KEY (falling back to OPENAI_API_KEY, or an explicit -k for the session).
3. Code with DeepSeek
# general coding — DeepSeek V4
agentty --provider deepseek -m deepseek-v4-pro
# the thinking model — reasoning streams live as it works
agentty --provider deepseek -m deepseek-reasoner--provider and -m persist. Switch provider live with ^P and model with ^/ — both have a search filter, so type deepseek to jump to it. With deepseek-reasoner, the model's chain-of-thought streams live in agentty's reasoning view, the same place Claude/Grok/Gemini thinking appears.
Why DeepSeek + agentty
- Very cheap, very capable — frontier coding at a fraction of the cost.
- Native tool-calling — read/edit files, ripgrep search, shell, apply-patch, and subagents all work through standard OpenAI-style function calls.
- Live reasoning for
deepseek-reasoner, streamed as it thinks. - No lock-in — switch to Kimi, Claude, GPT, or Gemini live, no restart.
See the providers docs for details, or the deep-dive: Kimi K2 and DeepSeek in agentty.