Help & Guides
Your guide to understanding and using the Alkimi platform.
Knowledge
Knowledge is the foundation of intelligence in Alkimi. It allows you to transform your agents from generic assistants into specialized experts grounded in your proprietary data. By organizing documents, websites, and text into structured Collections, you provide the definitive source of truth that your agents use to answer questions, reason through problems, and automate tasks.
The Knowledge section is your central hub for managing these assets. Here, you can ingest diverse data sources (from PDF manuals and research papers to live documentation websites) and organize them into secure, permission-controlled libraries. Once a Collection is connected to an agent, it acts as a private, long-term memory that the agent can query instantly, ensuring every response is accurate, relevant, and strictly based on the information you've verified.
Creating a Collection
To get started, you'll need to create a collection. This acts as a container for your documents. You can have multiple collections for different topics or projects.
Create Collection
Profile
Each Knowledge Collection has a profile page where you can manage its name and description. This page also shows you which agents currently have access to the collection, allowing you to track its usage across your organization.
Danger Zone
The profile page contains the "Danger Zone," where you can permanently delete the entire collection and all the files within it. This action cannot be undone.
Retrieval Strategy
A retrieval strategy determines how your agent sources information from its knowledge base. When you upload a document, it's broken down into smaller pieces of text called "snippets" or "chunks". When a user asks a question, a search is performed to find the most relevant snippets. The retrieval strategy you select is then applied to each of those matched snippets individually. For example, if you choose "Surrounding Snippets" and three snippets are matched, the agent will receive the neighboring context for all three.
Choosing the right strategy is key to balancing context relevance, response speed, and cost. It's important to remember that this is a strategy, not a guarantee. To prevent overloading the agent, the system may reduce the amount of included context if it exceeds the agent's available capacity.
The diagram below illustrates the basic flow, and the sections that follow explain each strategy in detail.
Relevant Snippets Only
Sends only the most relevant snippet.
Use Case: This is the most focused and cost-effective strategy. It's ideal for knowledge bases where context is not critical, such as a glossary, a dictionary, or a list of FAQs. The agent gets only the most direct piece of information, which is efficient but may lack broader context.
Surrounding Snippets
Sends the relevant snippet plus its neighbors.
Use Case: A balanced approach that provides immediate context. Useful for documents where the preceding and succeeding paragraphs help explain the matched snippet, such as a research paper, a legal document, or a news article.
Surrounding Pages
Sends the entire page containing the snippet.
Use Case: Provides broader context. Best for content where the entire page is needed to understand a single point, like a user manual where a diagram on the page is referenced in the text, or for technical documentation where context is spread across a page.
Entire Item
Sends the entire document if any part matches.
Use Case: Ensures complete context. Ideal for shorter documents where the full text is essential for understanding, such as a poem, a short story, a press release, or a legal contract where every clause could be relevant.
Always Include
Sends the entire document regardless of the user's query. Use sparingly.
Use Case: Best for critical, high-level documents that must always inform the agent's responses, such as a company's mission statement, a set of core safety protocols, or a brand style guide.
Permissions
Control who can use and manage this collection. Similar to agents, you can set a default access level for your organization and grant explicit, overriding permissions to specific members.
Collection Roles
Note: The roles listed below are the defaults for Knowledge Collections. Your organization's administrators can customize these permissions and create new roles to fit your specific needs.
Owner / Admin
Can fully configure the collection, manage its content, and control permissions.
- Use in Agents
- Read Content
- Manage Items
- Manage Settings
- Manage Permissions
- Delete Collection
Manager
Can fully configure the collection and manage its content, but not control permissions.
- Use in Agents
- Read Content
- Manage Items
- Manage Settings
- Manage Permissions
- Delete Collection
Editor
Can edit the collection's settings and manage its knowledge items.
- Use in Agents
- Read Content
- Manage Items
- Manage Settings
- Manage Permissions
- Delete Collection
Viewer
Can view the collection's settings and use it in agents.
- Use in Agents
- Read Content
- Manage Items
- Manage Settings
- Manage Permissions
- Delete Collection
User
Can use the collection's knowledge via agents.
- Use in Agents
- Read Content
- Manage Items
- Manage Settings
- Manage Permissions
- Delete Collection
Content Management
Content within a collection is organized in a hierarchical tree structure, similar to a file system. You can create folders to organize your knowledge sources.
Alkimi AI BETAKnowledge Configuration
It's important to note that the folder hierarchy has no bearing on agent responses. It is purely an organizational tool to help you manage permissions and activate or deactivate content in bulk.
Naming & Structure
To maintain a clean and functional structure, the following naming rules apply:
- Item names cannot contain a forward slash (
/). - Folder names cannot exceed 64 characters.
- The names for files, text, and website items cannot exceed 150 characters.
While the system supports nesting folders up to 32 levels deep, we recommend keeping your folder structure relatively shallow (2-3 levels deep) to ensure easy navigation and management.
Content Types
You can add content to your collections in several ways:
Add File(s)
Select files to upload.
Max 100MB & 1000 pages per file.
Auto: Automatically detects the best extraction mode for the file content. Recommended for most use cases.
Upload documents directly from your computer. Supported formats include PDF, DOCX, TXT, MD, and LaTeX. When uploading, you can select an Extract Mode to determine how the text is processed:
- Auto: (Recommended) Automatically detects the best extraction mode for the file type.
- Verbatim: Preserves the exact wording and structure. Ideal for legal documents or contracts where phrasing is critical.
- Distilled: Summarizes content to extract key information. Best for text-heavy or verbose documents where you want to capture the essence without the fluff.
Add Website
Enter the URL of the website to add.
Deep Crawl (Depth 2): The crawl will follow links up to 2 levels deep.
Only pages with this prefix will be crawled. Empty defaults to the starting URL.
The maximum number of pages to crawl (1-200).
Auto: Automatically detects the best extraction mode. Recommended for most URLs.
Create a separate item for each page, organized in a new folder.
Use an AI to generate additional content for each crawled page.
Ingest content from a website URL. This is a powerful way to keep your agent up-to-date with online documentation or company info.
Key Options:
- Crawl Depth: Controls how "deep" the crawler goes. A shallow crawl only processes the starting page and immediate links, while deeper crawls explore further.
- URL Prefix: Limits crawling to a specific section of the site (e.g.,
example.com/docs). Only pages starting with this prefix will be added. - Max Pages: Sets a hard limit on the number of pages to process to control credit usage.
- Extraction Mode: Same as with files, you can choose how content is processed (Auto, Verbatim, or Distilled).
- Processing Options:
- Single: Combines all crawled pages into one large knowledge item.
- Multiple (Subfolder): Creates a separate item for each page, organized in a new folder.
- Multiple (Inline): Creates separate items for each page in the current location.
- Content Generation: Optionally use AI to generate a synopsis or Table of Contents for each page to improve searchability.
Add Text
Enter the text content you want to add.
0 / 1,000,000
For quick snippets, notes, or content that doesn't exist in a file or URL, you can paste plain text directly into the system.
This is useful for:
- Copying sections from an email or chat.
- Adding specific instructions or context.
- Pasting code snippets.
Note: Text additions consume credits based on character count (1 credit per 25k characters).
Activating & Deactivating Content
Each item in your collection, including folders, can be individually activated or deactivated. Deactivating an item temporarily removes it from the knowledge base available to agents without permanently deleting it. This is useful for temporarily excluding certain documents or entire folders from agent responses.