Documents
Open in CMSThe Documents screen lists the files your agent uses as references when it answers, and is where you upload new ones. Found in the agent sidebar under Knowledge → Documents. Clicking a document opens its detail page.
Documents List
| New Document | Opens the Add New Document form. |
| Search Documents | Shows or hides the search panel below the buttons. |
| Sort By | Sorts the list by Code, Title or State, in Ascending or Descending order. |
| Grid View / List View | Switches between thumbnail cards and a compact list. Your choice is remembered. |
| Document card | Shows the thumbnail, a state badge, the title, the code, the author and how many chunks are active (for example "12 of 12 active chunks"). Click it to open the detail page. |
| State badge | Initial, Processing, Pending, Active, Inactive, Activating, Deactivating, Reactivating, Cancelled, Failed or Deleted. |
| Load more data | Loads the next page of documents. |
Search Documents
| Code | Finds documents by code. |
| Title | Finds documents by title. |
| Author | Limits the list to one registered author. |
| State | Limits the list to one state (for example Active or Failed). |
| Apply | Runs the search. |
| Reset | Clears every search field. |
Add New Document: File and Ingestion Mode
The form first asks for the file and how it should be processed. The remaining cards appear once a file is selected.
| File | The file to upload (required). It is stored temporarily and deleted after training finishes. |
| Document Guidelines | Expands tips for a good source file: plain text is preferred, the allowed file types are listed, remove page headers, footers and page numbers, and repeat table headers on each page. |
| Ingestion Mode | How the document is split into chunks (required). Options: Adaptive, Regulatory Context, Fact Sheet, Skimming, Semantic. |
| Mode Guide | Expands a short description of each ingestion mode. |
... button | Opens Ingestion Mode Parameters for the selected mode. If a parameter in there is invalid, an error appears under the field until you fix it. |
Ingestion Modes
| Adaptive | AI-powered. Analyzes the document structure and picks the best chunking approach. Use it when you are not sure of the document type. |
| Semantic | Basic. Splits the document into balanced sections while keeping the flow connected. The simplest choice when unsure. |
| Skimming | Structured. For long, hierarchical documents such as books, manuals or multi-chapter reports. Keeps chapters and sections, and can summarize sections that contain sub-sections. |
| Regulatory Context | Smart. For legal and compliance documents. Follows the legal structure (for example one article per section) and keeps related provisions together. |
| Fact Sheet | Smart. For short documents of about 1 to 3 pages. Keeps the original layout and captures charts with their explanations. |
Ingestion Mode Parameters
The fields depend on the selected mode. In create mode the dialog closes with Set Parameters; on a processed document it is read-only.
| Content Training AI Model | The model used for training and embedding generation (Adaptive, Skimming, Regulatory Context, Fact Sheet). For Regulatory Context, only models that accept PDF and TXT input are listed. |
| Content Size (in tokens) | Semantic only. How much text goes into each chunk. Smaller chunks (200 to 400 tokens) find precise answers, larger ones (600 to 1000) keep more context. |
| Overlapping Ratio (%) | Semantic only. How much content each chunk shares with the next one. Higher values give smoother flow but more chunks and cost. |
| Start Page / Finish Page | The page to start from and the last page to include. Leave Finish Page empty to process until the end. |
| Excluded Pages | Adaptive only. Types of content the AI should leave out when structuring the document. Type a custom entry and press Enter. |
| Use Summarization | Skimming only. Adds short summaries to higher-level sections. May add usage. |
| Keep Small Sections Separate | Skimming only. When on, small sections are not merged into larger ones. |
| Minimum / Maximum Section Size (in tokens) | Skimming only. When small sections are merged, and how large a section can get. |
| Use Full LLM Enrichment | Adaptive only (Advanced Parameters). On: every section goes through the LLM, which may improve retrieval but costs more and takes longer. Off: only sections that reference others are sent. |
| LLM Enrichment Sections Threshold | Adaptive only. The most sections sent to LLM enrichment. Leave empty for no cap. |
| Minimum / Maximum Chunk Size (in tokens) | Adaptive only. Size hints passed to the AI when it structures the document. |
| Page Overlap | Adaptive and Skimming. Pages repeated when a long document is split, to keep context. |
| Document Type Hint | Adaptive only. Optional hint such as "product manual". Leave empty for auto-detection. |
| Table as Plain Text | Semantic and Skimming. Treats tables as plain text. Turn it off only for documents with many complex tables. |
| Skip Image Analysis | Semantic and Skimming. When on, images are ignored, which saves time and cost. When off, images are described by AI. |
| Image Description AI Model | The model that describes images. Always shown for Adaptive; for Semantic and Skimming only when Skip Image Analysis is off. |
Document Details
| Title | The main name of the document (required). |
| Author | A registered author, shown as code and name (required). Authors are managed under Citations → Authors. |
| Code | Your own unique identifier for the document (required), for example P02. |
| Thumbnail Source | Choose URL (paste an HTTPS image link and click Preview) or Upload (pick an image file). |
| Document Source | A link to the original source or reference location. |
| Permissions | Permission codes that restrict this document to authorized users only. |
| Save / Cancel | Save creates the document and starts processing. If references are turned off for the agent, you are asked whether to enable them. |
Document Detail
| Document State | The current state of the document and the number of active chunks. |
| View Chunks | Opens Chunks filtered to this document. |
| Chunking Process | The chunking status, with a View Log button. |
| View Log | Opens the Chunking Log with a refresh button and an Auto Refresh option while processing runs. |
| Document Preview | Shows the uploaded file for PDF, TXT and Markdown documents. |
| Editing | Title, Author, Code, thumbnail, Document Source and Permissions can only be edited while the document is Inactive. |
Detail Actions
The buttons in the page header depend on the document's state.
| Deactivate | Shown when the document is Active. Stops the agent from using it. |
| Activate | Shown when the document is Inactive. Puts it back into use. |
| Retrain | Shown when Inactive. Lets you choose a new ingestion mode and parameters, then Retrain replaces all existing chunks. |
| Retrain as New | Available in retrain mode. Copies the document and processes the copy as a new document. The Code must differ from the original. |
| Delete | Shown when Inactive. Deletes the document after confirmation. |
| Cancel Chunking | Shown while the document is processing. Stops the processing. |
| Retry Chunking | Shown when processing was Cancelled or Failed. You confirm by typing the document title. |
| Purge Document | Shown for Deleted, Failed or Cancelled documents. Removes it permanently after you type the document title. |
Related guide: Documents