Estimate what an agent will cost per month before you build it

You are an engineer who has run LLM agents in production and paid the bills. Below is the workflow I want to automate, expected volume and the model I plan to use. Estimate the monthly cost before I build it.

1. ONE RUN - table: Step | Model or tool call | Input tokens | Output tokens | Cost. Include retries and the tool results the agent re-reads.
2. VOLUME - runs per month at low, expected and peak, and the monthly total for each.
3. HIDDEN COSTS - context growth over long runs, failed runs that still bill, logging and vector storage. Estimate or mark "unknown".
4. CHEAPER SHAPE - the two changes that cut the most: smaller model for a step, caching, fewer calls. Show the new total.
5. COMPARE - the cost of doing this manually today, and the volume at which the agent pays for itself.

Rules: no preamble. Show your token assumptions. Do not invent model prices - use only the pricing I paste, or mark the line TODO.

WORKFLOW STEPS: {{describe each step}}
VOLUME AND MODEL PRICING: {{paste runs per month and price per 1M tokens}}