/slash FAQ
1. What is Slash?
Token Optimization for Context Engineers. You do your job exactly the same with less tokens. Simple as that.
2. How does it work?
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.
- a) Try it —
bunx slash-tokens - b) Install it —
npm install slash-tokens - c) Import it —
import 'slash-tokens/auto'
3. What does it cost?
Try it now — anyone, no account:
Have a key? View the data.
bunx slash-tokens. Then two options.
Solo — real email, we send a key, $20 on the house. That's the account. Month 1: saved you $100, for free. Month 6: $880 cumulative — or lifetime. We show it. We don't charge. Get a key.
Team — $39 / month. Save even more on Annual: $390 / year ($32.50 / month). $39 for the data. We enrich and compile reports. Dashboard to view it. On demand. Richer over time.
Enterprise — custom. Talk to us.
Example this month: $1,000 savings. Slash 10% = $100 waived. Team $39 for the data. Additional savings $61.Team — $39 / month. Save even more on Annual: $390 / year ($32.50 / month). $39 for the data. We enrich and compile reports. Dashboard to view it. On demand. Richer over time.
Enterprise — custom. Talk to us.
Have a key? View the data.
4. Can I cancel? Am I committed, tied in with a subscription?
bunx — stop running it and you're out. Solo is not a subscription. Team is Stripe monthly or yearly — cancel anytime; bunx keeps running.
Deep Dive
5. How does Slash prevent a call?
Through the hosted Slash proxy (with a key), the Gate checks every call before it reaches the provider:
- Duplicates — same content sent twice within 60 seconds? Stopped.
- Empty/trivial — nothing useful to send? Stopped.
- Context overflow — tokens exceed the model's window? It would fail anyway. Stopped.
- Rapid fire — a burst of calls in 10 seconds? Throttled.
slash-tokens/auto) checks and routes each call on your machine, but never blocks one.
PREVENT isn't about blocking legitimate work. It catches the accidental stuff — the redundant calls, the "oops I hit send twice," the requests that were never going to work anyway.
6. Which is most common — Prevent, Route, or Pass?
ROUTE is the workhorse. Most API calls don't need the flagship model. A simple question on Opus costs 5x more than Haiku — same answer. Slash catches those and reroutes automatically.
PREVENT is surgical — catches obvious mistakes and redundancies.
PASS lets the right call through unchanged. No savings, no charge. Right tool for the job.
PREVENT is surgical — catches obvious mistakes and redundancies.
PASS lets the right call through unchanged. No savings, no charge. Right tool for the job.
7. Is this per project or do I turn it on once?
Turn it on. Forget about it. Every call optimized. One switch. Fire and forget. The Gate runs sub-millisecond on every request — whether it's one prompt or thousands a day.
8. Is it safe? What happens to my API keys?
Your API keys pass straight through to the provider. Slash never stores them. It only makes routing decisions. 4.8 KB WASM, zero dependencies, MIT licensed, open source. You can audit the code yourself at GitHub.
9. Who built this?
FAF — the team behind the IANA-registered Foundational AI-Context Format (
application/vnd.faf+yaml). Founded by wolfejam. Part of an ecosystem with 120,000+ downloads across 18 packages and 3 registries. Launched April 2026.
10. Which providers and models are supported?
Anthropic (Fable, Opus, Sonnet, Haiku), OpenAI (GPT-6 Astra, Sol, Luna; GPT-5.6; GPT-5.4), xAI (Grok 4.7, 4.6, 4.3), Google (Gemini 3.1 Pro, 3.x Flash, Flash-Lite), and NVIDIA Nemotron on Nebius Token Factory (Ultra, Super, Nano, Lightning). 48 models with built-in pricing;
slash-tokens/auto intercepts the first four providers. Don't see yours? Open an issue.