What has changed
Until now, you had to think of uploading an extra document yourself in the chat — the original contract behind a payment reminder, say, or the notice behind an objection. From now on the AI takes care of that: beneath every answer, small “Add a document” cards appear whenever a specific document would clearly improve the assessment.
How it works
- Cards appear in context — for a question about a payment reminder, for instance: “Do you have the underlying contract?” For an official notice: “Do you have the proof of service?”
- Upload straight from the chat — PDF, JPG, PNG or WEBP. Hard limit: 5 MB and 10 pages per document. That keeps token costs and answer quality in an honest balance.
- A maximum of 5 submissions per chat, 20 per day — this guards against endless uploads and keeps the system usable for everyone.
- A transparent price before you upload — the card shows exactly, in advance: “1 credit base + 1 per page”. You confirm before anything is charged. If the text cannot be extracted, you are refunded automatically.
- The document immediately becomes active context for the conversation. Your next question refers to it automatically — no need to enter anything twice.
Why not “free”?
We briefly tried it that way, but we had to be honest: OCR, text extraction and recalling the document into every follow-up AI answer cost tokens — and therefore money. We would rather charge small, transparent amounts for a concrete added benefit than hide them somewhere in a subscription. If you would rather not pay the price, you can put the question in a new chat instead — that too is now possible with a single click from the follow-up bar.
Why this matters
Most legal answers stand or fall on context. A payment reminder without the original contract is reading the tea leaves; an objection without the notice is flying blind. Until now, many users left the chat at exactly the point where the relevant document was sitting on the desk — but no one had explicitly asked for it.
→ Try it in the chat: /app/chat · What the AI can’t do: /system-grenzen
