Admin Dashboard

Checking…

API Cost every call counted — answers, agentic steps, embeddings, email AI, background jobs

By component

By model

Calls include background jobs (email scans, event extraction, indexing) — so calls won't match question counts, and cost can accrue on days with zero questions. Cached repeat answers log $0. Fixed infrastructure (AWS + Supabase) is separate: ≈ $125/mo — see the notes at the bottom.

Mode Comparison runs your schedule under all three data architectures and shows exactly what differs

From To

Pick a date range (1–3 days is fastest) and hit Run. The server answers the same schedule question three times — once per mode — and lists every item that one mode found but another didn't, with how long each took.

Question Comparisons every question asked with Comparison mode on, answered by all three architectures

Loading…

Usage by User

User	Questions	Tokens	Est. Cost	Activity

Feedback Reports

Data Indexing

Loading jobs…

Daily Breakdown

Date	Questions	Tokens	Est. Cost

Cost Breakdown Notes

What the API Cost panel tracks: every OpenAI and Gemini call — final answers, agentic-loop steps, routing, embeddings, email AI, and background jobs — priced per token (cached input tokens at the 90% discount). Prices from OpenAI Pricing. The ledger starts the day this shipped; older calls weren't recorded, so "All Tracked" undercounts history. The per-user table below still counts only each question's final answer call.

Not tracked above (fixed monthly infrastructure — June 2026 actuals):

AWS ≈ $62/mo — app server $18, VPC + load balancer + DNS + ECR ~$44 (+$32/mo once the sync worker + Redis land). Check AWS Billing Console.
Supabase ≈ $25/mo — Med Caddy's share of the Pro org (the other ~$10 on the invoice is the Piri project). Check Supabase Dashboard.
Google APIs — Calendar, Gmail, Drive, Tasks are free under standard quotas.

Current models (July 2026): answers & planning on gpt-5.4-mini ($0.75 / $4.50 per 1M tokens in/out); optional gpt-5.4 answers ($2.50 / $15). Email classifier also gpt-5.4-mini. Typical agentic-loop question ≈ 1–2¢; full inbox sweep ≈ 3–5¢.

Tip: OpenAI costs scale with users. AWS/Supabase are mostly fixed. At pilot scale (150 students, 1–2 questions/day each), expect roughly $70–150/mo in OpenAI costs on gpt-5.4-mini.