Wikidoc

📚 wikidoc

Sort your documents into a corpus you can actually query.

wikidoc files your documents, sorts them by what they contain, and writes down what each one is. Once that's done, you stop digging through folders, you ask.

🔍 What people use it for

• 🧾 Answering a tax claim. The tax office says a payment was never credited — pull the transfer proof, the return it belongs to, and the dates, in one question.
• 📊 Feeding the accountant. Every document for one company and one fiscal year, already in the right folder, before they have to ask twice.
• 🏠 Property paperwork. Rent receipts, charge statements, insurance, management reports — see what exists and what's missing.
• 📮 Forwarding receipts. Supplier invoices found and sent to your accounting inbox without opening a folder.
• 🧠 Giving an assistant real context. An LLM that knows which company existed when, who you worked with, and what you decided — because it's all written down.

🚀 First run

No config yet? It sets itself up:

1. Checks what your machine can do.
2. Reads a sample of your documents: company IDs, dates, who keeps sending you things.
3. Asks you only what it couldn't work out on its own: "this company ID shows up in 47 files — whose is it?"
4. Writes your config, in plain YAML you can read and edit.
5. Dry run: shows what it would move, moves nothing.

About ten minutes, once.

⚙️ How a pass works

Text gets extracted from your files first. Then a router sorts each document into one of three lanes:

• route — a rule recognised it. The agent double-checks and files it.
• propose — sensitive, or unclear. The agent opens the file before doing anything.
• residual — a scan, or genuinely ambiguous. Vision agents read it properly.

Rules live in your `config.yaml`. After each pass the agent suggests new ones from what it just learned, and you decide whether to keep them — so the same kind of document doesn't need thinking about twice.

📦 What go