What you get on day 7 — BrainOfBrains

From: John @ BrainOfBrains · To: Sam

Your 7-day KostAI report — one leak, $127/mo

📅 Sent on day 7 🔍 Watched 4,312 calls 💡 One clear fix

Hi Sam,

TL;DR — the number you were hoping for

KostAI watched 4,312 calls over the last week. One specific pattern is costing you about $127/month right now. The fix is five lines of code. We walk through it below.

Projected annual savings from the single fix

$1,524 / year

At your current call volume. Grows if usage grows.

Where your AI spend went this week

📚

Sending a whole book when a paragraph would do

Your chat endpoint re-sends the full conversation history every call. Average call carries 12,400 tokens of history that wasn't needed.

$127/mo

🧠

Using the expensive brain for easy work

12% of your calls are simple classifications routed to GPT-4. A smaller model would have been fine.

$34/mo

🔁

Asking the same thing twice in a minute

Your rate-limit retry logic is firing three times on failure. Seen 28 times this week.

$9/mo

✅

Everything else

The other six categories were clean. Your cache strategy is solid.

healthy

The fix for the big one

In api/chat.ts, trim the history you pass into the model. Five lines, zero risk to response quality on the sample we tested:

// Before:
const response = await openai.chat.completions.create({
  model: "gpt-4o",
  messages: conversationHistory // sends everything
});

// After:
const trimmed = conversationHistory.slice(-6); // last 6 turns only
const response = await openai.chat.completions.create({
  model: "gpt-4o",
  messages: trimmed
});

We verified on 50 random calls from your logs that response quality is unchanged with 6 turns of history vs. full. If you want, we can run a bigger quality check before you ship — just reply.

After you ship the fix, here's what to expect

Your input token count per call drops by ~70% on the chat endpoint.
Expected bill impact: ~$127/month lower starting day one.
KostAI keeps watching. If the fix doesn't land cleanly, we'll see and tell you.
If you want, schedule another 7-day check in a month to confirm the savings held.

Nice work running a clean AI setup overall — six of nine leak categories came back healthy, which is better than most stacks we see. The one big fix above is the only thing worth your engineer's time this week.

Happy to jump on a quick call if any of this is unclear.
— John, BrainOfBrains

This is a real template, but the name, numbers, and code have been anonymized and rounded. Your report will be tailored to what KostAI actually finds in your stack.

What you get, seven days after install.

Ready for one of these in your inbox?