Skip to content
Sırdaş
es · Español
Try it freeToolsAI datasets NewDevelopersServidor MCPCLIPythonn8nTypeScriptAgent SkillSkill: invoicesSkill: contractsBlogBusiness

New · Context engine in your browser

Ask a folder. Give your AI only what it needs.

Drop a folder of PDF, Word, Excel, email or transcript files and ask a question. Sırdaş finds the passages that answer it, with their page and evidence, and hands you a context of a few thousand tokens ready to paste into ChatGPT, Claude or Gemini. Your files never leave this tab.

Drop a folder or several files here

They are read in your browser and forgotten when you close the tab. Nothing is uploaded or stored.

PDF · Word · Excel · PowerPoint · OpenDocument · EPUB · HTML · CSV · .eml email · text · .srt and .vtt subtitles · scans and photos with optional OCR

Sent to servers: 0 BNothing left your device

How it works

  1. 01

    Reads and splits

    Each file becomes text with its pages and sections, in a Web Worker. A file identical to another is recognized by its fingerprint and not read twice.

  2. 02

    Searches and picks

    Ranks passages by the words in your question (BM25) and fills the token cap with the best ones. Ask about a date or an amount and it also brings the passage that holds it.

  3. 03

    Hands it over, labelled

    Every passage comes with its file, page and the line that justifies it, wrapped as data: the AI reads it, it does not obey it.

What it does and what it doesn't

  • Matches words, ignoring accents and plurals: “contracts” finds “contract”
  • Cites file, page, section and the exact line
  • Hides names, IDs, emails and phones with the same placeholder across every file
  • Warns you when a PDF hides text a person cannot see
  • It does not answer the question: it gives you the context, and your AI writes the answer
  • It does not understand synonyms: “car” does not find “vehicle”
  • It does not read scans or photos unless you ask: OCR is optional, takes seconds per page and can make mistakes
  • It keeps nothing: close the tab and it is forgotten

The same, for your agent

The MCP server and the CLI use this same engine over whole folders on your machine, with per-file memory and measured token counts on every call. Available since version 0.4.0.

For developersHow to check nothing leaves

Questions

Are my files uploaded anywhere?
No. They are read inside your browser, in a Web Worker, and live in this tab's memory. The meter below counts any request the page makes; you can also check the Network tab in DevTools.
What is it for if it does not answer?
So you do not paste 300 pages into the AI. You give it only the passages that matter, with their page, and it answers better, faster and cheaper. And what you do not give it, it does not see.
Why does it sometimes miss the answer?
It searches by words. If the question and the document say the same thing with different words, it can miss. Use the words the document would use (“expires” rather than “ends”).
How many files can it take?
Up to 500 at a time and 100 MB per file. With many large documents it depends on your device's memory.
What about scans and photos?
They are set aside and wait: you decide whether to read them with OCR. It runs in your browser with models hosted on this same site, in Spanish and English, and takes a few seconds per page. A skewed or blurry scan can produce errors, which is why every citation carries its page so you can check it.