# Document Inbox > Drag in a whole folder of client documents, drop a zip, or pick many files at once. The AI reads each one, recognises what it is, and files it to the right company — anything it can’t place waits for a one-click assign. _Updated 2026-07-11_ Source: /help/document-inbox **In short:** The Document Inbox (in Data Governance → Document Inbox) turns a pile of files into filed, recognised records. Drop a folder, a .zip, or many files; CorpSec AI shows a cost estimate, reads each file, classifies it, and files the ones it recognises to their company. The rest land in Pending Review, where the AI’s best guess is pre-filled so you just click Assign. It never guesses-and-files silently. > **Note:** Reading scanned PDFs and images uses AI vision, which is metered. The inbox always shows an honest cost estimate before you commit, and files that exceed your quota still upload and store — only the AI reading pass is skipped. Nothing is charged if billing is not configured. ## What the Document Inbox is for Corporate secretaries accumulate documents — passports, ACRA BizFiles, signed resolutions, constitutions, KYC evidence — often as a messy folder or a zip from a client. The Document Inbox is the fast lane for getting all of it into CorpSec AI: instead of opening each file, working out whose it is, and filing it by hand, you drop the whole lot in and let the AI sort it. Find it under Data Governance → Document Inbox. It has two parts stacked together: the bulk ingest drop zone at the top, and the Pending Review queue below it. _[screenshot: The Document Inbox: a large "Drag a folder or files here" drop zone with Choose folder / Choose files buttons, and a Pending Review list below.]_ ## Three ways to bring documents in All three intake modes converge on the same queue, so mix and match as you like: - Drag a whole folder in from your desktop — CorpSec AI walks it recursively, sub-folders and all. - Choose folder to pick a directory, or Choose files to multi-select individual files. - Drop a .zip — it is unpacked on the server and its contents counted after unpacking. > **Note:** Folder drag-and-drop is a desktop feature. On a phone, use Choose files or your camera — you can snap a passport or BizFile straight in. See Mobile app. ## Check the cost estimate first (live) Before you commit, CorpSec AI shows exactly what the batch will cost. Each file that needs AI vision (a PDF or image) uses one extraction; text, Word, CSV and spreadsheet files are read for free. The estimate shows how many files are queued, how many need vision, and — if metering is on — how many extractions you have left this month on your plan. _[screenshot: The queued-files view: a cost-estimate card ("12 files queued · 5 need AI vision (≈5 extractions) · 7 free"), a "These all belong to" company selector, and an Import button.]_ > **Heads up:** If a batch would exceed your remaining extractions, CorpSec AI tells you plainly: the files over the cap still upload and store, only the AI vision reading is skipped. You can top up next cycle or upgrade — nothing is lost. > **Note:** If no billing is configured, extraction metering is off — nothing is capped or charged, and the estimate says so. ## Optionally tell it who they belong to If you already know every file in this batch belongs to one company, pick it from the These all belong to selector — otherwise leave it on Auto-detect per file (recommended) and let the AI work out each file’s owner from its contents. ## What happens when you import Click Import and CorpSec AI processes the batch — a progress bar tracks it, and large batches are handled in chunks so nothing is dropped. When it finishes you get a summary with five numbers: _[screenshot: The "Import complete" summary: five stat tiles (Files, Filed to company, Pending review, Duplicates, Degraded) and a per-file list showing each file’s recognised type and where it was filed.]_ | Result | What it means | | --- | --- | | Files | Total processed in this batch. | | Filed to company | Confidently recognised and attached to their company — done. | | Pending review | Couldn’t be confidently matched — waiting for you below. | | Duplicates | Already on file — the stored original was reused, not duplicated. | | Degraded | Read with limitations (e.g. an unreadable scan) — flagged so you can check. | > **Note:** Files that contain personal data (a passport, an NRIC) are marked Sensitive in the result list, and personal identifiers are masked wherever they are shown. See Security & your data. ## Pending Review — the honest tail (live) Any document the AI could not confidently attribute waits in Pending Review rather than being filed to a guess. For each row you see the file name, its recognised document type, and the AI’s best subject guess with a confidence percentage. Where that guess matches a company you already have, it is pre-selected — so filing it is a single click. _[screenshot: The Pending Review queue: rows with a document-type chip, an "AI guess … % confident" line, a company dropdown pre-filled in green, and a Confirm/Assign button.]_ 1. **Review the AI’s guess** — Each row shows "AI guess: Acme Pte. Ltd. · 78% confident" and whether it matched a company on file. 2. **Confirm or pick the company** — If the guess is right and pre-filled, click Confirm. Otherwise choose the correct company from the dropdown and click Assign. 3. **Assign several at once** — Tick multiple rows and use the batch bar — "Assign all selected to…" — to file a whole group to one company in one go. 4. **Download the original** — The download icon on each row fetches the authorised original file if you need to check it before assigning. > **Tip:** Rows that failed a field check (for example a passport whose expiry date could not be validated) are flagged "N fields need a look" so you know to open the original before filing. ## Where recognised documents show up Once a file is filed to a company, what the AI read from it appears on the Insights tab of the right panel (under More) whenever that company or a related document is selected — the document type, classification confidence, the fields it checked, an AI summary, and an integrity fingerprint. You can also drop a file straight into a task conversation and it is ingested the same way. > **Tip:** Dropping files into the centre-column chat is covered in AI chat and memory. ## Frequently asked questions ### What file types can I drop in? PDFs and images (which are read with AI vision), plus text, Word, CSV and spreadsheet files (which are read for free). A .zip of any of these works too — it is unpacked on the server. ### Will it create duplicate records if I import the same folder twice? No. Files already on record are detected as duplicates and the stored original is reused. Re-dropping the same folder is safe. ### What happens to files it can’t read? They still upload and store — nothing is thrown away — and they are marked "degraded" with the reason, so you can review or re-supply a cleaner copy. ### Does it ever file a document to the wrong company automatically? It only auto-files when it is confident. Anything uncertain goes to Pending Review for a human to confirm. It never guesses-and-files silently. ### Do these extractions cost money? Vision reading of PDFs/images is metered against your plan when billing is enabled; text and Office files are free. The inbox shows the estimate before you commit, and over-quota files still upload — only the AI reading is skipped. ## Related - [Data governance](/help/data-governance) - [AI chat and memory](/help/ai-chat-and-memory) - [Companies](/help/companies) - [Security & your data](/help/security-and-data) - [Import your client book](/help/import-client-book)