InfraCost

**Current:** Claude Max plan at €180/mo (~$196/mo) — upgraded from Pro €18/mo around Mar 6. **Alternative:** Downgrade to Pro €18/mo + rely on API credits for heavy usage. **PROS:** - Saves up to €162/mo (~$178/mo, $2,136/yr) — largest single cost item - Pro plan sufficient for light conversational use - API usage ($65/mo) already covers programmatic needs **CONS:** - Max plan provides 5x higher rate limits and priority access - May hit rate limits on Pro during intensive development sessions - Extra Usage credits (€50/mo) would no longer be available - Development velocity could decrease if hitting API limits **VERDICT:** Only worth downgrading if development pace slows. Monitor actual Max usage patterns first.

**Current:** Each project has prod + test web service + prod + test DB = 4x cost. Example: homegrif has homegrif_com ($7.13) + homegrif_com-test ($7.13) + homegrif-db ($6.42) + homegrif-db-test ($10.70) = $31.38/mo **PROS:** - Could save $20-40/mo by sharing test DBs or using Neon branching (free tier) - Neon already has homegrif-neon and scrabsnap-neon instances - Test DBs don't need persistence or high availability **CONS:** - Shared test DBs can cause data conflicts between developers - Neon branching has cold-start latency (~1-2s) - Migration effort per project - Need to ensure test data isolation **ACTION:** For projects with Neon instances, migrate test environments to Neon branching. Keep prod DBs on Render.

**Current:** 6 databases provisioned at 15GB disk (~$10.70/mo each). **Target:** Recreate with 1GB disk (~$6.42/mo each). **PROS:** - Saves ~$25.68/mo ($308/yr, ~€23.62/mo) - Biggest single optimization available - No performance impact (actual data uses <100MB per DB) **CONS:** - Cannot resize in-place — must create new DB, migrate data, update connection strings - Requires downtime per DB (~15min each, 6 DBs = ~90min total) - Risk of missed connection string updates breaking services - DBs affected: homegrif-db-test, oncoteam-db-prod, oncoteam-db-test, partners-db-prod, partners-db-test, scrabsnap-db

**Current:** partners-cz-prod using Standard plan ($0.0336/hr) for part of the month + Starter ($0.0094/hr) for the rest — ~$13.18/mo projected. **Target:** Ensure it stays on Starter plan ($0.0094/hr) — ~$7.13/mo. **PROS:** - Saves ~$6/mo ($72/yr) - Starter tier sufficient for current traffic levels - No migration required — just plan change **CONS:** - Standard tier provides 1 vCPU + 2GB RAM (vs 0.5 vCPU + 512MB on Starter) - May have been upgraded for a reason (traffic spike, memory pressure) - Need to verify current RAM/CPU utilization before downgrading **ACTION:** Check Render metrics for partners-cz-prod. If P95 memory <400MB, safe to lock on Starter.

**Current:** 476/500 free minutes used at day 14 (95.2%). Overage: $5/1000 min. **Risk:** Will exceed 500 min this month. New services (budgetco, contacts-refiner, oncoteam) driving more builds. **PROS:** - Avoids $5-10/mo overage charges - Forces build discipline (fewer unnecessary deploys) - Can enable "auto-deploy on push" selectively **CONS:** - Restricting builds slows development iteration - May need to manually trigger deploys - Alternative: accept small overage as cost of faster iteration **ACTION:** Disable auto-deploy on test environments. Use manual deploys for non-critical services. Consider merging PRs less frequently.

**Current:** infracost on Render free web + Basic-256mb DB ($6.42/mo). **Target:** Railway Hobby plan ($5/mo) already running — add infracost there. **PROS:** - Saves ~$1.42/mo on DB (Railway DB may be cheaper or free within plan) - Reduces Render pipeline pressure (fewer build minutes consumed) - Consolidates infrastructure monitoring tool on a second platform (reduces single-vendor risk) - Railway Hobby plan already being paid for **CONS:** - Migration effort: need to set up Railway service, migrate DB, update DNS - Railway Hobby plan has limits (8GB RAM, 8 vCPU total across all services) - May conflict with oncofiles and oncoteam resources on same plan - Render free web tier is actually $0 — only the DB costs money **VERDICT:** Small savings but good for pipeline pressure. Defer until pipeline overage becomes recurring.

**Current:** robota ($0.21) and robota-test ($0.21) show as "Deleted" but still charged in March billing. **PROS:** - Saves $0.42/mo (minor) - Cleans up billing noise **CONS:** - None — these are already deleted services - Charges should stop automatically next billing cycle **ACTION:** Verify in April billing that charges stopped. If not, contact Render support.

**Current:** $27.98 credits remaining as of Mar 14. Auto-reload is DISABLED. **Risk:** Credits will run out in ~2 weeks at current burn rate ($65/mo), causing API outage for all services using Claude API. **PROS:** - Prevents unexpected API outage - Zero cost (auto-reload just adds credits, doesn't increase spend) - Required for production reliability **CONS:** - Enables automatic spending (could spike if usage anomaly) - No built-in spend cap on Anthropic platform - Need to monitor monthly usage manually **ACTION:** Enable auto-reload with $50 threshold. Add monthly usage alert.