[OpenCode Field Notes Part 3] Provider strategy to reduce token/subscription interruptions

1. Why provider strategy matters

As discussed in Part 1, my main problem was not raw model quality. It was workflow interruption in long sessions.

For heavy LLM usage, the pattern is simple: once limits hit mid-task, focus breaks and productivity drops.

So I prioritize operating strategy before model benchmarks.

2. My real provider setup

This section is based on my actual ~/.config/opencode/opencode.json setup.

• Plugins: opencode-antigravity-auth@latest, opencode-anthropic-auth, opencode-cursor-auth
• Provider blocks: google, anthropic, github-copilot, opencode, cursor
• Additional channel: ollama

My most-used practical flow looked like this:

• Google-auth-based Gemini Pro family
• Antigravity-auth-based Claude Sonnet family
• GitHub Copilot Education as a support channel
• Switching between existing subscriptions/auth paths by task condition

The key was flexibility inside one OpenCode environment, not dependence on one provider. Compared to jumping across web tools, this made my CLI routine much more stable.

3. Problems I actually saw and how I switched

Claude Sonnet quality was strong, but on heavy days tokens drained fast. So I shifted to a direct handoff pattern whenever limits appeared.

Typical sequence:

1. Implement/refactor on primary model
2. Switch to backup model at first sign of limit pressure
3. Move repetitive work to cost-efficient models

This turned “hard stop” into “channel switch,” which kept momentum.

4. Operating principles I follow

• Keep complex implementation on high-quality primary models
• Offload long log scans/summaries/repetitive checks to support models
• Prepare at least two auth/subscription channels to distribute risk
• Stay within each service’s terms and school policy boundaries

The goal is not one perfect model. It is a full-day development loop that does not break.

5. What changed in practice

• Less anxiety around token/subscription limits
• Faster switching than waiting for recovery
• Better focus retention in long sessions
• Less manual handoff work when starting new sessions

Result: more depth on actual tasks for the day.

6. Next

In the next post, I will show how I built parallel sub-agent workflows with oh-my-opencode.

Sources

• opencode-antigravity-auth: https://github.com/NoeFabris/opencode-antigravity-auth
• opencode-cursor-auth: https://github.com/POSO-PocketSolutions/opencode-cursor-auth
• opencode-anthropic-auth (npm): https://www.npmjs.com/package/opencode-anthropic-auth

Note: opencode-anthropic-auth shows a deprecation notice on npm. Check current status before production use.