> For the complete documentation index, see [llms.txt](https://docs.helvia.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.helvia.ai/knowledge/knowledge-base.md). # Knowledge Base A Knowledge Base (KB) is a collection of articles, made from the documents and content you bring in. That content can come from files you upload, articles you write by hand, or external systems. Knowledge Bases are how you ground your agents in your own content: at runtime, the agent retrieves the most relevant articles and writes an answer from them (RAG). Go to **Workspace > Knowledge Bases** to manage the Knowledge Base and the articles inside it.

### How Knowledge Is Organized Knowledge is the body of content your agent can draw on to answer a question: product documentation, policies, FAQs, internal guides, anything you want it to reference. The platform structures that content so the agent can pinpoint the right section on demand, rather than scanning a whole document each time. ```mermaid flowchart LR S1["📄 Manual Upload"] --> KB S2["☁️ Knowledge Base
Integration"] --> KB KB["📚 Knowledge Base"] --> G["🗂 Groups"] G --> A["📝 Articles
tagged and segmented"] A --> SN["🔍 Semantic Search Node"] SN --> AG["🤖 Agent"] style KB fill:#eeedfc,stroke:#615DEC,color:#1a1a2e style G fill:#eeedfc,stroke:#615DEC,color:#1a1a2e style A fill:#eeedfc,stroke:#615DEC,color:#1a1a2e style SN fill:#fff4d6,stroke:#F3A702,color:#1a1a2e style AG fill:#fff4d6,stroke:#F3A702,color:#1a1a2e style S1 fill:#f0f4ff,stroke:#94a3b8,color:#1a1a2e style S2 fill:#f0f4ff,stroke:#94a3b8,color:#1a1a2e ``` A Knowledge Base is organized in three layers. The Knowledge Base itself holds one or more Groups, each Group holds Articles, and Articles are the units an agent actually retrieves at runtime. An Article is rarely a one-to-one copy of its source file: the platform segments long documents during import so each one is small enough to be retrieved precisely, and large enough to carry useful context. {% hint style="info" %} **Retrieval at Runtime** The articles retrieval, ranking, and tag filtering at runtime are handled by the Semantic Search node on the agent workflow. {% endhint %} ### Creating a Knowledge Base When you create a new Knowledge Base, you decide how to fill it: * Bring the content in yourself * Connect an external system that brings it in for you. Both options start from **Add New KB** on the Knowledge Bases page. #### Internal / Manual Upload Internal Knowledge Bases live entirely inside the platform. You add content by uploading files one at a time, importing a CSV in bulk, or writing articles directly in the built-in editor. Everything is editable in place and there is no ongoing sync, so you stay in full control of every article. Pick this path for curated FAQs, internal SOPs, or any material you do not have stored in another system. #### Integration Knowledge Base integrations connect to an external content system you already use and pull the content into the platform for you. Once the connection is set up, ingestion happens automatically and the articles stay in step with the source. You can also trigger a manual sync from the Knowledge Bases table whenever you want to refresh the content on demand. Pick this path for large repositories you do not want to copy by hand. See the [Integrations](/administration/integrations.md#knowledge-base-1) page for the supported connectors and how to set each one up. #### Processing Options When an integration processes a source file, or when you upload one manually, the platform parses, segments, and indexes it. The following options control how that happens.


Process Mode	Standard for fast rule-based parsing, or AI-Powered for complex layouts and tables
Max Article Size	Pick a chunk size from Small (~750 characters) to XLarge (~9,000) to balance retrieval precision against context
Include Images	Extract images for the agent to reference in its replies (PDF sources only)

Process Mode

Standard for fast rule-based parsing, or AI-Powered for complex layouts and tables

Max Article Size

Pick a chunk size from Small (~750 characters) to XLarge (~9,000) to balance retrieval precision against context

Include Images

Extract images for the agent to reference in its replies (PDF sources only)

### AI-Powered Processing Documents rarely come in clean. Tables span multiple pages, scans hide behind images, layouts mix columns and call-outs. AI-Powered processing reads every page with visual language models and segments by meaning. Articles preserve the structure of the source instead of breaking at fixed character counts. This AI semantic chunking is the biggest lever on answer quality: when an article holds one complete idea rather than a fragment cut mid-thought, the agent retrieves cleaner context and answers more accurately. ```mermaid flowchart LR D["📄 Source Document"] --> O["🔎 OCR + Visual Language Model"] O --> C["✂️ Semantic Chunker"] I["💬 Custom Instructions
(optional)"] -.-> C C --> A["📝 Articles"] style D fill:#f0f4ff,stroke:#94a3b8,color:#1a1a2e style O fill:#eeedfc,stroke:#615DEC,color:#1a1a2e style C fill:#eeedfc,stroke:#615DEC,color:#1a1a2e style A fill:#eeedfc,stroke:#615DEC,color:#1a1a2e style I fill:#fff4d6,stroke:#F3A702,color:#1a1a2e ```


Visual Language Model OCR	Reads any document type the way a person reads a page, not as a stream of disconnected tokens
Semantic Chunking	Splits the source at topic boundaries instead of character limits, so each article carries one coherent idea
Custom Instructions	Add natural-language guidance for the segmenter directly on the Upload File modal: keep tables intact, group by section heading, ignore footers
Built for Real Documents	Available for PDF, DOCX, PPTX and HTML. Handles complex layouts that trip up rule-based parsing

#### Choosing a Processing Mode The two modes differ in how they read a source file, so match the mode to the document: * **Standard:** Fast, rule-based parsing. The right default for plain-prose sources with simple layouts, where it is quick and accurate enough. * **AI-Powered:** Reads each page with a visual language model and segments by meaning. Reach for it when the source has tables, scans, or complex layouts, or when you need custom instructions for the segmenter. If Standard processing mode leaves you with broken structure or merged tables, reprocess that source with AI-Powered mode. ### The Knowledge Bases Table The table at **Workspace > Knowledge Bases** is where every Knowledge Base lives. Select any row to open that Knowledge Base in the editor and work on its articles or settings.

A few actions live here: * **Add New KB:** Start a new Knowledge Base * **Manual Sync:** Refresh an integration-backed Knowledge Base from its source on demand. Disabled for Internal KBs * **Delete:** Remove a Knowledge Base and unlink it from every agent that uses it * **Bulk Actions:** Tick rows and act on several Knowledge Bases at once * **Search in Contents:** Find a Knowledge Base by its name ### Working with Articles Articles are the building blocks of a Knowledge Base. Each one is a small, self-contained piece of content the agent can retrieve when it needs to answer a question.


Retrieval Units	The smallest piece of content an agent retrieves at runtime
Rich-Text Editing	Write or refine articles inline with formatting controls
Grouped	Sort related articles into groups inside a Knowledge Base
Tagged	Add tags for retrieval scoping and easier housekeeping
Publishable	Toggle articles in or out of agent retrieval without deleting them
Source-Traced	Link each article back to the document or URL it came from

Everything you do with articles happens on the **Articles** tab of an open Knowledge Base. The tab is split into a navigation tree, with Groups and their articles, and a rich-text editor for the article you currently have selected.

#### Adding Articles You can add articles to a Knowledge Base in three ways: * **New Article:** Write an article by hand. Opens an empty rich-text editor. * **Upload File:** Upload a single source document. The file is parsed and split into one or more articles based on the processing options. See [Supported File Types](#supported-file-types) for accepted formats. * **Import CSV:** Bulk-add or update many articles at once by [uploading a CSV](#import-and-export). The dialog provides a downloadable sample template. The Import action lives on the **Settings** tab of the Knowledge Base. #### Supported File Types Upload File accepts a single document up to 50 MB. The file type determines whether AI-Powered processing applies or the file falls back to Standard parsing. | File type | Standard | AI-Powered | Notes | | --------- | -------------------- | -------------------- | --------------------------------------- | | PDF | :white\_check\_mark: | :white\_check\_mark: | Source page number tracked per article | | DOCX | :white\_check\_mark: | :white\_check\_mark: | - | | HTML | :white\_check\_mark: | :white\_check\_mark: | - | | PPTX | :white\_check\_mark: | :white\_check\_mark: | - | | XLSX | :white\_check\_mark: | :x: | Each worksheet becomes a markdown table | | TXT | :white\_check\_mark: | :x: | - | | MD | :white\_check\_mark: | :x: | - | #### Organizing with Groups Groups are folders that organize related articles inside a Knowledge Base. They are purely organizational: retrieval still happens across the whole KB regardless of which Group an article sits in. Use them to keep the editor tidy, especially in Knowledge Bases that hold hundreds of articles.

To create a Group, select **New Group** and name it. Move articles between Groups through the **Group** field in the article editor. #### The Article Editor Selecting any article (or creating a new one) opens the editor on the right side of the tab. The editor is rich-text so you can write or refine an article inline without exporting elsewhere. The Details panel carries the per-article settings: * **Published:** Toggle the article in or out of agent retrieval. Unpublished articles stay editable but are skipped at runtime, so you can take an article offline without deleting it. * **Group:** Move the article into a different Group * **Tags:** Apply or remove tags from the article. Articles synced from a Knowledge Base integration inherit the tags set on that integration. * **Source URL:** Where the article came from. Auto-filled by Knowledge Base integrations as a read-only link, editable for manually created articles * **Page Number:** The page where the article begins in the source document (PDF only) {% hint style="info" %} The editor does not autosave. Use **Save Changes** to keep your edits. {% endhint %} #### Searching Articles The **Articles** tab has its own search bar. Use the dropdown to pick a search mode:

#### Deleting Articles To remove an article from a Knowledge Base, open it in the editor and select **Delete Article** at the bottom of the page. The article is removed immediately and is not recoverable, so export the Knowledge Base first if you may need the content again. ### Configuring a Knowledge Base The **Settings** tab is where you adjust a Knowledge Base after creation. Use it to rename the KB, refine its description, or expand its language coverage.

#### Import and Export Two CSV-based actions are available at the **Settings** tab: {% columns %} {% column %} #### Import Bulk-add or update articles by uploading a CSV. Max file size 5 MB. {% endcolumn %} {% column %} #### Export Download every article in the Knowledge Base as a CSV file, for backup, migration, or offline review. {% endcolumn %} {% endcolumns %}

Sample import CSV format

Each language is a pair of columns, `[LANG].Title` and `[LANG].Body`. Add another language by adding another pair. ```csv EN.Title,EN.Body,ES.Title,ES.Body,Tags,Status Reset your password,"Open **Settings**, then select [Reset password](https://example.com).",Restablecer tu contraseña,"Abre **Ajustes** y selecciona [Restablecer contraseña](https://example.com).",account,published Business hours,"We are open 9 to 5, Monday to Friday.",Horario de atención,"Estamos abiertos de 9 a 5, de lunes a viernes.",general,published ``` * **Language codes:** Use the ISO 639-1 two-letter code, case-insensitive (`EN`, `ES`, `EL`, `DE`) * **Primary language:** The Knowledge Base's primary language must include both its Title and Body columns * **Row-level columns:** `Tags` (comma-separated) and `Status` apply to the whole row. Optional `[LANG].GroupId` and `[LANG].GroupName` columns sort articles into groups. {% hint style="info" %} The Import dialog includes a downloadable sample CSV. Use it as the basis for your file so your columns match the structure Import expects. {% endhint %}

### Deleting a Knowledge Base To remove a Knowledge Base, open it and select **Delete Knowledge Base** at the bottom of the **Settings** tab. The KB, every article inside it, and every agent link are removed immediately. {% hint style="danger" %} **Permanent Action** Deleting a Knowledge Base removes every article and unlinks it from every agent that uses it. Export the content first if you may need it again. {% endhint %} ### Best Practices * **Start with Standard processing:** Use Standard parsing first and switch a source to AI-Powered only when you see broken structure or merged tables in the resulting articles * **Match article size to your content:** Pick Small for short FAQs where precision matters, and Large or XLarge for dense reference material where retrieval should pull surrounding context * **Tag from day one:** Apply a consistent tag scheme as you ingest content. Retro-tagging hundreds of articles later is painful. {% hint style="success" %} You now know how to organize content into Knowledge Bases, choose between manual and integration ingestion, tune how source files are segmented, and find the right KB inside the Workspace. {% endhint %} --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.helvia.ai/knowledge/knowledge-base.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.

:wand-magic-sparkles:

Process Mode

:ruler:

Max Article Size

:image:

Include Images

:eyes:

Visual Language Model OCR

:scissors:

Semantic Chunking

:comment-pen:

Custom Instructions

:files:

Built for Real Documents

:cube:

Retrieval Units

:pen-to-square:

Rich-Text Editing

:folder-tree:

Grouped

:tags:

Tagged

:eye:

Publishable

:link:

Source-Traced