savemytokens

Pre-flight triage for AI agents — know the cheapest fetch path before the crawl burns tokens.

Try it liveHow to use it

Run triage

Paste a URL. Get a recommended action — llms.txt, raw fetch, headless, or abort on WAF.

Loading…

Usage

Call triage before you fetch. Then follow the recommended action instead of defaulting to a full HTML crawl or headless browser.

REST

curl "https://savemytokens-api.vercel.app/v1/triage?url=https://example.com"

MCP

Point Cursor or Claude Desktop at the local stdio server, then ask the agent to call triage_url before fetching a page. MCP setup guide.

Why it matters

Building a small feature often means an agent opens many docs and package pages. Triage first: use llms.txt when it exists, raw GET when HTML is enough, and abort on WAF instead of stuffing challenge pages into context. Run pnpm ab:triage locally for a live with/without session estimate (bytes → tokens heuristic — not measured LLM billing).

Follow the action

  • USE_LLMS_TXT — fetch the discovered Markdown; skip crawling HTML.
  • FETCH_RAW — plain HTTP GET is enough; no headless browser.
  • WAF_BLOCKED — abort or route to an unblocker before burning proxy reputation.

About

savemytokens is a pre-flight probe for AI web agents. Before navigation, it inspects a target URL and returns the cheapest, most token-efficient way to fetch content.

Calls GET /v1/triage on https://savemytokens-api.vercel.app · probe engine: @savemytokens/core