No Claude in Hong Kong? OpenCode Is the Safest Alternative (Gemini 3.5 Flash Is Best Value)

The key point: OpenCode is completely free and open source!
It also supports Ollama, so you can run models locally and offline. Your code never leaves your machine, which means top-notch security 🔒
🧰 4 must-try features
1️⃣ /init: generate an “AI project manual” in one click 📄 (a must for teams)
The first thing to do in a new project is type /init. OpenCode automatically scans your tech stack and coding conventions, then generates an agent.md in the root directory. Once you have this “AI manual”:\

• The AI’s answers fit your project 100%\
• You can commit it to Git so the whole team uses it\
• Onboarding new people gets much faster
  2️⃣ Plan Mode: think it through first, then get to work 🗺️
  It analyzes the structure and comes up with a complete plan. Only once you confirm it looks fine do you switch back to Build Mode to execute.
  That way you don’t have to worry about the AI changing files at random and leaving you to clean up the mess.
  3️⃣ small_model: saves you a big chunk of API costs 💰
  Remember to set this up in opencode.json:
  main_model (e.g. Claude / GPT‑4)
  small_model (DeepSeek / a Mini model)
  Simple tasks (summaries, titles, formatting) automatically use the cheaper model, and only complex tasks use the main model. Same quality, much lower cost.
  4️⃣ Local mode: the best fit for company environments 🔒
  It supports Ollama, which means:\
• Fully offline\
• No code uploaded\
• No worries about company policy\
• Safe and stable
  Many companies in Hong Kong don’t allow cloud AI, so this is a real lifesaver.
  🤣 One last question for everyone
  Which AI coding tool are you mainly using right now? Has anyone tried OpenCode’s Custom Tools yet? 👇