Context

We've added OpenAI and Google Gemini keys to opencode-zen. Both providers offer complex pricing:

• OpenAI: Free tier with daily token allowance under data-sharing agreement, then paid
• Google Gemini: Similar free tier deal with limits
• OpenCode: Always free
• Claude Max / Copilot: Subscription-based with monthly limits

Problem

Current budget orchestrator treats providers as either subscription or paid-per-token. Need hybrid support:

• Free tier until daily/monthly limit reached
• Automatic fallback when free tier exhausted
• Preference for free providers when available

Requirements

1. New Provider Types

• hybrid-free-then-paid: Free tier until limit, then paid
• hybrid-free-daily: Daily free reset

2. Free Tier Tracking

• Detect when limits reached (parse 429 errors, user config, heuristics)
• Track usage against free tier limits
• Reset counters at appropriate intervals
• Fallback logic when exhausted

3. Provider Selection Priority

1. Free tier available (opencode, openai/google within limits)
2. Subscription with headroom (Claude Max, Copilot)
3. Paid-per-token (when budget allows)

4. Parallel Agent Optimization

Distribute across free tiers for more parallel agents without cost.

5. Files to Modify

• src/features/budget-orchestrator/
• src/features/usage-tracker/
• src/config/schema.ts

Success Criteria

• System uses free tiers before paid
• Automatic fallback when exhausted
• Cost savings measurable
• More parallel agents possible
• All tests pass