Make any link, image or PDF readable for AI
Paste a link or drop a file and get clean, structured text back — even from scans and screenshots. Then split it into bite-sized pieces, ready for a knowledge base.
- 1 Choose
- 2 Read
- 3 Split
Sign in to use this toy
It runs on your AIVAX account, and each document is charged to your balance at list prices. You choose how long access lasts and can revoke it in the console.
No account? Create one for free, then come back.
or read a web page
Your document, as text
How it works
Read
AIVAX opens the page or file and pulls out its text, keeping the structure — headings, lists, tables. Web pages lose their menus and ads; images and scanned pages get read with OCR. Fetch & OCR
Split
A model finds where one idea ends and the next begins, so each chunk can answer a question on its own. Your text is never rewritten. Text segmentation
Search
Load the chunks into a collection and your assistant can find and quote the right passage of your content. Collections
For developers API calls from this session, and the code to build it yourself
Calls made by this page
- API calls
- 0
- Processing units
- 0
Estimates use public list prices before daily allowances. Each call is billed to the AIVAX account you signed in with, through an application key limited to web fetch and generations.
Build it yourself
Two API calls: POST /api/v1/web/fetch and POST /api/v1/generations/segment.
See the API reference or the
Worker behind this page.
# 1. Media → Markdown. Send links as they are and inline files as base64 data URIs # with their real MIME type (50 MB max per item). One call can mix several items. curl https://direct.inference.aivax.net/api/v1/web/fetch \ -H "Authorization: Bearer $AIVAX_API_KEY" \ -H "Content-Type: application/json" \ -d "{ \"contents\": [ \"https://example.com/article\", \"data:application/pdf;base64,$(base64 -w0 manual.pdf)\", \"data:image/png;base64,$(base64 -w0 receipt.png)\" ], \"returnErrors\": true }" # → { "data": { "results": [ { # "index": 0, # "extractedText": "# Title\n\nParagraph...", ← Markdown # "processingUnits": 7, ← what you pay for # "jsonProcessingUnits": 0, # "error": null }, ... ] } } ← one result per item, in order # 2. Markdown → segments. `sanitize` skips fragments that are useless for retrieval. curl https://direct.inference.aivax.net/api/v1/generations/segment \ -H "Authorization: Bearer $AIVAX_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "documents": ["# Title\n\nParagraph..."], "sanitize": true }' # → { "data": { # "result": [ { "index": 0, "count": 8, "segments": ["...", "..."] } ], # "usage": { "processing_units": 2516, "cost": 0 } } }
// Works in Node 18+, Bun, Deno and browsers. Keep the API key server-side. const AIVAX = "https://direct.inference.aivax.net/api/v1"; const headers = { authorization: `Bearer ${process.env.AIVAX_API_KEY}`, "content-type": "application/json", }; // 1. Media → Markdown. `content` is a link or a data URI. async function toMarkdown(content) { const response = await fetch(`${AIVAX}/web/fetch`, { method: "POST", headers, body: JSON.stringify({ contents: [content], returnErrors: true, }), }); const { data } = await response.json(); const [item] = data.results; // one result per `contents` entry, matched by index if (item.error) throw new Error(item.error); return { markdown: item.extractedText, processingUnits: item.processingUnits }; } // 2. Markdown → RAG segments async function segmentForRag(markdown, { sanitize = true } = {}) { const response = await fetch(`${AIVAX}/generations/segment`, { method: "POST", headers, body: JSON.stringify({ documents: [markdown], sanitize }), }); const { data } = await response.json(); return { segments: data.result[0].segments, processingUnits: data.usage.processing_units }; } // 3. Shape the segments as JSONL for POST /api/v1/collections/{id}/documents function toJsonl(segments, source) { return segments .map((text, index) => JSON.stringify({ docid: `${source}#${index + 1}`, // stable name: re-importing updates instead of duplicating text, __ref: source.slice(0, 64), // groups the chunks of one file __tags: ["aivax-toys"], __meta: { source, index }, })) .join("\n"); } // A web page is read straight from its link... const page = await toMarkdown("https://example.com/article"); // ...while PDFs and images go inline with their MIME type. const base64 = Buffer.from(await Bun.file("manual.pdf").arrayBuffer()).toString("base64"); const { markdown } = await toMarkdown(`data:application/pdf;base64,${base64}`); const { segments } = await segmentForRag(markdown); await Bun.write("manual.jsonl", toJsonl(segments, "manual.pdf"));
// Trimmed from src/worker.js in the repository. The real file adds the AIVAX OAuth // sign-in that stores each visitor's application key in an HttpOnly cookie. const AIVAX_API = "https://direct.inference.aivax.net/api/v1"; export default { async fetch(request, env) { const url = new URL(request.url); if (url.pathname === "/api/convert" && request.method === "POST") { const form = await request.formData(); const file = form.get("file"); const url = form.get("url"); const token = readCookie(request, "toys_session"); // visitor's application key if (!token) return Response.json({ error: "Sign in first." }, { status: 401 }); const content = url ? url // AIVAX opens the link itself : `data:${file.type};base64,${bytesToBase64(new Uint8Array(await file.arrayBuffer()))}`; const call = await callAivax(token, "/web/fetch", { contents: [content], returnErrors: true }); const item = call.data.results[0]; return Response.json({ markdown: item.extractedText, processingUnits: item.processingUnits, elapsedMs: call.elapsedMs, requestId: call.requestId, }); } return env.ASSETS.fetch(request); // static site from ./public }, }; async function callAivax(token, path, body) { const startedAt = Date.now(); const response = await fetch(AIVAX_API + path, { method: "POST", headers: { authorization: `Bearer ${token}`, // billed to the visitor's account "content-type": "application/json", }, body: JSON.stringify(body), }); const payload = await response.json(); return { data: payload.data, elapsedMs: Date.now() - startedAt, requestId: response.headers.get("x-request-id"), }; }