TL;DR: I built FAF to fix Context. Senior devs shrugged. This time I'm doing Tokens. This time your boss is listening.
The Pattern
Two years ago I built FAF — a way to give AI context about your project. IANA registered the format. 50,000 downloads across three registries. It works.
But here's what I learned: context is philosophical. Tokens are invoices.
By the time you find out a call was a miss, you already paid.
The Numbers
I evaluated the top AI repos on GitHub. Combined annual token sunk cost:
From 8 public repos. Production is 10-100x that.
And none of them are checking token cost before the call goes out. Not one.
The Fix: One Line
import 'slash-tokens/auto' Every LLM API call — Anthropic, OpenAI, xAI, Google — evaluated pre-call. Routes to the cheapest model that can handle it. Same answer. Less cost.
- Opus task? Stays on Opus.
- Haiku task on Opus? Routed to Haiku. 80% cheaper.
- Unnecessary call? Aborted. 100% saved.
4.8 KB. Sub-millisecond. Zero dependencies. You don't change how you work. Slash changes what you pay.
Try It
bunx slash-tokens Run it on any project. See the number. Show your boss.
How It Works
Evaluates every API call before it goes out. Salvages tokens you didn't need to send. Aborts unnecessary calls. Routes to a cheaper model when one fits.
Don't go to the corner shop in a Ferrari.
What It Costs
10% of what you save. You keep 90%. That's it.
Day one, every day. Never gonna change. Solo to SpaceX. Same deal.
The Numbers
- v1.1.1 — Released April 8, 2026
- 99/99 — Tests passing
- 10 — Models with built-in pricing
- 4.8 KB — Zig-compiled WASM
- Sub-ms — Evaluation time
Who's First?
The leaderboard is all empty today. Every company. Every repo. Nobody's optimizing pre-call.
Someone's going to be first. Someone's going to tell the story:
"We adopted /slash. Here's what we saved in month one."