GoogleCloudPlatform Cost Estimation
- Per-job cost: Use estimate_job_cost to show users the expected cost before submitting a prediction
- Supports AF2 monomer, AF2 multimer, and OF3 job types
- Returns per-phase breakdown (MSA, predict, relax) with both on-demand and DWS FLEX_START (spot) pricing side by side
- GPU auto-selection matches the submission tools (L4 for small proteins, A100 for larger)
- IMPORTANT: Always present both pricing columns so users can compare. FoldRun defaults to FLEX_START.
- Proactive use: When showing the pre-submission confirmation table, include cost estimate
- Monthly projection: Use estimate_monthly_cost for budget planning and TCO analysis
- Input monthly job volumes by type (AF2 monomer, multimer, OF3)
- Returns compute + infrastructure + other costs with monthly and annual totals for both pricing modes
- Infrastructure includes: Filestore, GCS, Artifact Registry, Agent Runtime, Cloud Run, VPC/NAT
- Use include_infrastructure=False to see compute-only costs
- Actual costs: Use get_actual_job_costs to retrieve costs for completed prediction jobs
- Calculates costs from actual Agent Platform job runtimes and machine specs
- Groups costs by pipeline run with per-phase breakdown (MSA, predict, relax)
- Shows estimate vs actual comparison with accuracy percentage
- Provide a pipeline_job_id for a specific run, or omit to see all recent jobs
- Use this when users ask "how much did that job cost?" or "show me my spending"
- IMPORTANT disclaimer: These tools return estimates based on Google Cloud list pricing. Always remind users that actual costs depend on their organization's pricing agreement with Google Cloud (negotiated rates, CUDs, enterprise discounts). The disclaimer is included in every tool response — surface it to the user.