Prepare your brand book and sources
Choose documents that answer brand questions: guidelines, company introductions, product descriptions, customer research or interview notes. Identify which versions are current.
Prefer readable text, clear headings and specific examples. Image-only PDFs and long decks are harder to verify than documents with text layers. If important information only appears in an image, keep the image and its context.
Import data you have permission to use. Check for unnecessary personal information, passwords and client data before uploading. Client A's documents belong in client A's brand.
Import files and links
File upload accepts PDF, Word (.docx), PowerPoint (.pptx), Excel (.xlsx), Markdown, plain text and images (PNG, JPG, WebP). Select up to 10 files per batch. A PDF can be up to 64 MB and 200 pages, and one that averages more than about 250 KB per page is rejected. Every other file must be under 4 MB, except images: those up to 32 MB are resized automatically before upload. The interface reports files that do not fit. Only the brand owner or an admin can import.
- Select the correct brand. While the profile is empty, use Bắt đầu (Getting started) to upload files. Afterwards, open Kho dữ liệu (Data library) → Nhập tài liệu, which is the Import page.
- In Bắt đầu, add your files. On the Import page, paste text that has headings (at least 200 characters; convert a PDF to Markdown first) or enter the address of a public website. A page behind a login may not be readable.
- Track reading and processing. If a file exceeds limits, split it by chapter or export a smaller version.
- Review the profile and original source entries after processing.
Web import and ingesting text through MCP are different flows. Web import writes the extracted information straight into the profile. MCP ingestion (antiem_ingest_brand_document) takes text only (PDFs and images must be uploaded on the web), needs mcp:write and an owner or admin role, and uses credit. It works in two ways: sending Markdown saves a document and creates proposals for you to review in Xu hướng (Growth); confirming one field from a document that is already saved writes that field after a confirmation step. Follow progress with antiem_get_brand_document_status. The two flows do not share one confirmation process.
Documents added through MCP or chat are not kept by default: after 7 days they move to Đã xoá gần đây (Recently deleted), unless you choose Giữ lại lâu dài (keep permanently) or the AI archives them when it ingests. A document that is evidence for a profile field is left in place.
Track processing and read sources
Antiem reads documents, classifies them, extracts profile information and checks related data. Long documents or many images can take more time. Read the error for the failed step rather than repeatedly resubmitting everything.
After processing, open the document in Kho dữ liệu (Data library) or the profile's source area. Evidence is anchored to the extracted text. If that text is wrong, extraction and subsequent answers can also be wrong; compare it with the original file.
Check names, numbers, tables, claim conditions and negations carefully. For example, “does not use synthetic fragrance” must not become “does not use fragrance.” When correcting an error, identify both the incorrect passage and the correct information.
Manage, replace and delete documents
Use titles that identify the source, version and effective date. Import a new version, review differences and then decide whether to remove the old source. Importing a new file does not establish that all previous information is obsolete.
Deleting a document removes it and its related evidence from the active profile. Review the impact before deleting the only source for important information.
Deleted documents remain in Đã xoá gần đây (Recently deleted) for 30 days for recovery. Xoá vĩnh viễn (Permanently delete) immediately removes the retained copy, with no recovery through the interface. Restoring can reconnect suitable citations; check them if the profile changed while the document was in trash.
For missing sources, read Sources and confidence.