Skip to navigation
Local agentic coding with opencode
07.10.26
opencode talks to local models through providers (Ollama, LM Studio, llama.cpp) configured in opencode.json / ~/.config/opencode/opencode.json. What actually works with opencode: - Local, tool-capable: - Qwen3-14B — has tool calling + think/no-think toggle; fits ~10GB at Q4. - Qwen3-Coder-30B-A3B — MoE, agentic-coding tuned, fast; ~18GB at Q4. - Devstral-Small-24B — built for coding agents. - Local, not R1 distills as your main driver. - API route (easiest, cheap): DeepSeek official API (deepseek-chat = V3, deepseek-reasoner = R1). V3 is tool-capable and costs pennies — far better than any 14B local for this.
Reply
Anonymous
Information Epoch 1791551210
Make each program do one thing well.
Home
Notebook
Contact us
