Dokko User Manual

Part 0: What Is Dokko?

The Problem Dokko Solves

Every company has documents — policies, reports, contracts, manuals, product guides, meeting notes. But finding information in those documents is slow and frustrating. You search, you scroll, you read pages looking for one answer. And if your colleague wrote something two years ago, good luck finding it.

Dokko is one platform that lets you ask your documents anything, see when its own answers are weak, and take action through tools you connect to it.

Three capabilities work together to make this possible:

• Answers grounded in your documents. Ask a question in plain language; Dokko searches your uploaded files and writes an answer using them as the source, with citations to the exact passages it used. Answers stay accurate, traceable, and private to your organization.
• A built-in quality check on the AI itself. Dokko can review its own past conversations and flag responses that look inaccurate or off-topic, along with the documents that produced them. You see exactly what to fix, replace, or add to your knowledge base — closing the loop between AI quality and your content, so the system gets better the more you use it.
• Agents that take action in the outside world. Through configurable AI agents and MCP (Model Context Protocol) tool servers, Dokko goes beyond answering. Agents can query a database, call an API, run a calculation, look something up on the web, or trigger any external system you give them access to. This is what turns Dokko from a smart Q&A tool into an assistant that can actually get things done — interacting with the world beyond your documents.

The Three Core Ideas

1. Your Knowledge Base (the Filing Cabinet)

A Knowledge Base is where Dokko stores and indexes your documents so they become searchable. You upload your files — PDFs, Word documents, web pages — and Dokko processes them, and stores that understanding in a searchable indicies.

Think of it like a very smart filing cabinet. Instead of finding a document by its folder or filename, you find information by asking a question: “What is our return policy?” or “Who do I contact for IT issues?”

2. Chat (the Smart Colleague)

Chat is the conversation interface. You type a question, Dokko does an advanced search on your documents, and the AI (artificial intelligence) model writes an answer using those documents as its source. At the bottom of every answer, you can see Sources — links to the exact documents used.

The key thing to understand: Dokko’s AI only knows what you give it. It does not search the internet. It only answers based on the documents you have uploaded to it. This is intentional — it keeps answers accurate, relevant, and private to your organization.

3. Agents (the Automated Assistant)

An Agent goes beyond answering questions. Agents can take a sequence of actions: search documents, run calculations, look up external systems, and report back a complete result. Think of an agent as a project manager who receives a complex task, breaks it into steps, and executes each step automatically.

Who Uses Dokko: Your Role

Every person who uses Dokko has a role that determines what they can see and do. When you log in, your role is already set by your Organization Admin.

Not sure what your role is? Log in and look at the left sidebar. If you only see “My Chat”, you are a User. If you see sections like “Knowledge Base”, “Infrastructure”, or “Administration”, you have a higher-level role.

Quick-Start Checklist by Role

If you are an Organization Admin, follow this order:

1. Log in and explore the Dashboard (Part 1)
2. Add your AI provider API keys (Part 3.1)
3. Create LLM configurations (Part 3.2)
4. Upload documents or configure web crawling (Part 2)
5. Set up the chat widget (Part 5.1)
6. Invite your team members and assign roles (Part 8.1)
7. Test the chat (Part 4.1)

If you are a Content Admin, start here:

1. Log in (Part 1)
2. Upload documents (Part 2.2)
3. Create and organize repositories (Part 2.4)
4. Configure the chat widget (Part 5.1)
5. Test in My Chat (Part 4.1)

If you are a regular User:

• Read Part 1 (logging in) and Part 4 (using chat). That covers everything you need.

Part 1: Getting Started — Your First Login

1.1 Receiving an Invitation

When an Organization Admin adds you to Dokko, you receive an email with an invitation link. The email comes from your organization’s Dokko instance and contains a button labeled Accept Invitation or a link to set up your account.

Note: Check your spam folder if you don’t see the email within a few minutes.

Click the link. It will open a web page where you set your password.

1.2 Setting Your Password

On the registration page, you will see two fields:

1. New Password — choose a strong password (at least 8 characters, mixing letters, numbers, and symbols is recommended)
2. Confirm New Password — type the same password again

A password strength indicator shows you how secure your password is. Aim for “Strong.”

Click Save or Register. Your account is now active.

1.3 Logging In

Go to your organization’s Dokko web address in a browser. You will see the login page with three fields:

1. Organization Name — the name of your company’s Dokko workspace (your admin can tell you this exact name; it is case-sensitive)
2. Email — your work email address
3. Password — the password you just created Click Log In.

Forgot your password? Click the Forgot password? link on the login page. Enter your email address, and you will receive a password reset email.

1.4 The Dashboard — What You See First

After logging in, you land on the Dashboard. This is your overview page. Depending on your role, you may see:

• Active Sessions — how many chat conversations are happening right now across your organization
• Agents — a list of AI agents configured in your organization
• Document Sources — a summary of your connected document storage
• LLM Providers — which AI providers are configured and their status If you are a regular User, you may land directly on the My Chat page instead.

1.5 The Sidebar Navigation

On the left side of the screen is the sidebar — the main navigation menu. It is divided into groups:

You only see the groups your role allows. A regular User sees only My Chat. An Organization Admin sees everything.

The sidebar can be collapsed by clicking the arrow icon on its edge — useful on smaller screens. Click it again to expand.

At the bottom of the sidebar you will see your email address and a Logout button.

1.6 Account Settings

To change your password or manage your personal account:

1. Click Account Settings in the sidebar (usually near the bottom)
2. To change your password: fill in your current password and then your new password twice, then click Save
3. To delete your account: click the Delete Account button

Warning: Deleting your account is permanent and cannot be undone. You will lose access to all sessions and history associated with your account.

Part 2: Knowledge Base — Feeding Dokko Your Documents

Content Admins and Organization Admins: This section is primarily for you. Regular Users do not need to manage the Knowledge Base.

2.1 Understanding the Structure

Before uploading anything, it helps to understand how Dokko organizes your documents. Think of it as a library:

Document Source  (The Library Building)
└── Repository   (A Section of the Library, e.g., "Legal", "HR", "Product Docs")
    └── Files    (Individual Books and Documents)

• Document Source: The top-level storage location. Your organization has one document source (connected to AWS S3 cloud storage). It is created automatically when your organization registers. You rarely need to touch this.
• Repository: A named collection of documents within the document source. You can have many repositories — one for “HR Policies”, one for “Product Manuals”, one for “Customer Contracts”, etc. When someone uses the chat widget, they search within specific repositories.
• Files: The actual PDF, Word document, web page, or other file that contains your content. Why does this structure matter? When you give a user access to the chat, you can specify which repositories they can search. This means your sales team only finds sales documents, HR only sees HR policies, and so on. Good organization leads to better, more focused answers.

2.2 File Management — Uploading Documents

Navigate to Knowledge Base > File Management in the sidebar.

This is the file browser — your document upload center. It shows a folder tree on the left and the contents of the current folder on the right.

Uploading Files

1. Navigate to the folder where you want to upload (or stay at the root)
2. Either:

• Drag and drop files from your computer directly onto the file browser area
• Click the Upload button and browse your computer to select files

3. You can upload multiple files at once
4. A progress indicator shows how many files have been uploaded

Supported File Types

Dokko can read and index: PDF files, Microsoft Word documents (.docx), Microsoft PowerPoint files (.pptx), plain text files (.txt), HTML pages, Markdown files (.md), and more.

Creating Folders

1. Right-click in the file browser area, or look for a New Folder button
2. Type a name for the folder
3. Press Enter to confirm Use folders to organize related documents together — for example, a folder called “2024 Policies” containing all policy PDFs from that year.

Renaming and Deleting Files

Right-click any file or folder to open the context menu. From there you can:

• Rename — change the file or folder name
• Delete — remove the file permanently

Warning: Deleting a file removes it from Dokko’s storage. This cannot be undone. Make sure you have the original file elsewhere before deleting.

Marking a Folder as a Repository

Once you have organized documents into a folder, you can “promote” that folder to a Repository so the AI can search it.

1. Right-click the folder in the file browser
2. Select Mark as Repository from the context menu
3. Give the repository a name and description
4. Click Save The folder is now a searchable repository that you can assign to chat widgets and user permissions.

2.3 Document Sources

Organization Admins only: Regular users and Content Admins do not need to configure document sources.

Navigate to Knowledge Base > Sources in the sidebar.

A Document Source is the cloud storage system that powers your organization’s document library. When you first registered, Dokko automatically created one for you connected to AWS S3 (Amazon’s cloud storage service). You do not need to create another one.

However, you can adjust how the AI processes and searches your documents through the document source settings.

Configuration Options Explained Simply

Parsing Strategy — how the AI “reads” your documents when you upload them:

Chunking Strategy — how the AI slices your documents into searchable pieces. Think of chunking as deciding how big each “quote” from your documents can be:

Contextual Expansion — when enabled, each chunk is placed within the broader context of the document it came from, so its meaning stays clear even when read on its own. This noticeably improves semantic retrieval, especially for short or ambiguous passages. Recommended for most knowledge bases; there is a small one-time processing cost during indexing.

Hybrid Search — when enabled, Dokko uses two search methods at once: keyword matching (like a standard search engine) AND meaning-based search (finding relevant information even if you use different words). Enable this for better results in most cases.

Reranking — after finding initial matches in your documents, a second AI model re-evaluates and re-scores them for relevance. This improves answer accuracy but adds a small amount of processing time.

Note: After changing document source settings, documents that were already uploaded may need to be re-indexed to benefit from the new settings.

2.4 Repositories — Organizing Your Knowledge

Navigate to Knowledge Base > Repositories in the sidebar.

A Repository is a named collection of documents the AI can search. You create repositories to organize your knowledge into logical groups.

Creating a Repository Manually

1. Click Create Repository (or the + button)
2. Fill in:

• Name: a clear, descriptive name (e.g., “HR Policies 2024”, “Product Documentation”, “Customer FAQs”)
• Description: optional — a short note about what this repository contains

3. Click Save After creating the repository, add documents to it either by uploading files through File Management and marking the folder as this repository, or by using the web crawler (see below).

Web Crawling — Importing Website Content Automatically

If your knowledge lives on a website or intranet (an internal company website), Dokko can automatically read and import those pages for you. This is called web crawling.

Navigate into a repository, then look for the Crawler or Web Crawler section.

Setting up a crawl:

1. Starting URL: the web address where the crawler begins (e.g., https://help.yourcompany.com)
2. Maximum Depth: how many clicks deep the crawler will follow links

• Depth 1 = only the starting page
• Depth 2 = the starting page and every page it links to
• Depth 3 = one level deeper still
• Start with 2 or 3; too deep can import thousands of irrelevant pages
• Set to -1 if you want no depth limit

3. Include URL Patterns: optionally restrict the crawler to only pages whose addresses contain specific text (e.g., /docs/ to only crawl your documentation section)
4. Exclude URL Patterns: skip pages matching certain patterns (e.g., /blog/ to skip blog posts)
5. Click Start Crawl The crawler runs in the background. You can check its status:

Scheduled Re-Crawling

If the website changes regularly, you can schedule Dokko to re-crawl it automatically:

• Recrawl Frequency: how often (in hours) to run the crawl again
• Recrawl Starting Date: when to begin the first automatic recrawl This keeps your knowledge base up to date without any manual work.

Part 3: Infrastructure — Setting Up the AI Engine

Organization Admins and Content Admins: This section covers the AI model configuration. Think of it as choosing the engine and fuel for your car.

Regular Users can skip this section entirely.

3.1 LLM API Keys

Navigate to Infrastructure > LLM API Keys in the sidebar.

What Is an API Key?

To use an AI model like ChatGPT (from OpenAI) or Claude (from Anthropic), Dokko needs permission to use that service on your behalf. That permission comes in the form of an API key (Application Programming Interface key) — essentially a private password tied to your account with the AI provider.

When Dokko uses the AI to answer questions, the usage is billed to your account with that provider. The API key is how the provider knows who to charge.

Supported AI Providers

How to Get an API Key

1. Go to the provider’s website and create an account (e.g., platform.openai.com for OpenAI, console.anthropic.com for Anthropic)
2. Navigate to their API Keys section (usually under Settings or Account)
3. Click Create New Key or similar
4. Copy the key immediately — most providers only show it once

Adding an API Key to Dokko

1. Navigate to Infrastructure > LLM API Keys
2. Click Add Key or Create
3. Fill in:

• Name: a label so you remember what this key is for (e.g., “OpenAI Production Key”, “Anthropic Main”)
• Provider: select the AI company from the dropdown
• API Key: paste the key you copied from the provider’s website

4. Click Save

Security note: Dokko stores your API key securely encrypted. It is never displayed in plain text after saving.

3.2 LLM Configurations

Navigate to Infrastructure > LLM Configurations in the sidebar.

What Is an LLM Configuration?

A configuration (short for LLM Configuration, where LLM means “Large Language Model” — the AI brain) is a recipe that tells Dokko:

• Which AI model to use
• How the AI should behave (precise or creative)
• How long answers can be

You can create multiple configurations for different purposes: one for the public-facing chat widget (precise, formal), another for internal brainstorming (more creative), another for document processing (very precise).

Creating an LLM Configuration

1. Click Create on the LLM Configurations page
2. Fill in the fields:

1. Optionally attach a Prompt Template (see Part 3.3)
2. Click Save

Tip on Temperature: For a company knowledge base assistant that should give accurate, grounded answers, use a temperature of 0.0 to 0.3. Higher temperatures are better for creative writing or brainstorming.

Setting a Default Configuration

One configuration can be marked as the Default. This fallback is used whenever no specific configuration is specified for a chat widget. It is good practice to set a reliable, well-tested configuration as your default.

3.3 Prompt Templates

Navigate to Infrastructure > Prompt Templates in the sidebar.

What Is a Prompt Template?

A Prompt Template is a background instruction you give to the AI — before the user even types anything. The user never sees this instruction, but it shapes every response the AI gives.

Examples of what a prompt template can do:

• Set a persona: “You are a friendly HR assistant for Acme Corp. Always refer to the company as ‘we’.”
• Restrict topics: “Only answer questions about our products. If asked about anything else, politely decline.”
• Set a language style: “Always respond in formal, professional English. Never use slang.”
• Set the language: “Always respond in French, regardless of the language the user writes in.”

Creating a Prompt Template

1. Click Create on the Prompt Templates page
2. Give it a Name (e.g., “Formal English HR Assistant”)
3. Select the Type:

• Chat: used for user-facing conversations
• Preprocessing: used internally when the AI processes documents

4. Write your instructions in the text editor
5. Click Save Once saved, you can attach the template to an LLM Configuration (in the configuration’s settings).

3.4 Configuration Sets

Navigate to Infrastructure > Configuration Sets in the sidebar.

A Configuration Set bundles multiple LLM configurations together under one name. For example, you might create a set called “Standard Deployment” that includes:

• A Chat configuration (for answering questions)
• A Preprocessing configuration (for processing documents)

When you connect a configuration set to a document source or chat widget, all the right configurations are applied automatically. This saves you from having to specify each configuration separately every time.

3.5 RAG Settings

Navigate to Infrastructure > RAG Settings in the sidebar.

What Does RAG Mean?

RAG stands for Retrieval-Augmented Generation. In plain English: before the AI writes an answer, it first searches your documents to find relevant information. Then it uses that retrieved information as the factual basis for its response.

This is what makes Dokko’s answers grounded in your actual documents, rather than the AI making things up.

The RAG Settings page lets you adjust the default retrieval behavior for your whole organization — how many document chunks to retrieve, whether to use reranking, and how results are scored. These defaults apply unless overridden per document source.

For most organizations, the default settings work well and no changes are needed here.

Part 4: Chat — Talking to Your Documents

4.1 My Chat

Navigate to My Chat in the sidebar.

This is the main chat interface for asking questions about your documents.

Starting a Conversation

1. Click inside the message input field at the bottom of the screen
2. Type your question in plain language — exactly as you would ask a colleague

• Example: “What is our policy on remote work?”
• Example: “Who is the contact person for IT support?”
• Example: “Summarize the key points from the Q3 report.”

3. Press Enter or click the Send button
4. The AI’s response will appear, streamed word-by-word in real time

Reading the Response

The AI’s answer appears in a chat bubble. At the bottom of the answer you may see a Sources section — these are the exact documents the AI used to formulate its answer. Click a source to open the original document.

Important: If the AI says “I don’t have information about that” or gives a vague answer, it means the relevant documents either haven’t been uploaded to Dokko yet, or you don’t have permission to access those repositories. Contact your Content Admin.

Continuing and Starting Conversations

Your conversation is automatically saved as a session. You can:

• Keep typing follow-up questions in the same session to continue the conversation
• The AI remembers the context of your current session (e.g., if you asked about “the leave policy” earlier, you can follow up with “how many days does it allow?” without repeating yourself)

4.2 Agent Run — Advanced AI Tasks

Navigate to Agent Management > Agent Run in the sidebar.

Content Admins and Organization Admins: Agents must be created before they can be run here (see Part 6).

What Is an Agent Run?

Regular chat answers one question at a time. An Agent can handle complex, multi-step tasks:

• Search multiple sources
• Perform calculations
• Look up external systems
• Combine results into a structured report

Think of it as the difference between asking a question and delegating a task.

Running an Agent

1. From the Agent Run page, select an agent from the dropdown menu
2. Type your request in the input field (this is your task description for the agent)
3. Click Send
4. Watch the agent work in real time:

• You will see function calls appear — these are the steps the agent is taking (e.g., “Searching documents…”, “Looking up web results…”)
• Each function call shows its result
• Finally, the agent writes a complete response

Stopping an Agent

If the agent is taking too long or going in the wrong direction, click the Abort button to stop it immediately.

Part 5: Interfaces — Deploying Chat to Your Website or App

Content Admins and Organization Admins: This section covers deploying Dokko’s chat and search capabilities to your website or internal tools.

5.1 Chat Widget

Navigate to Interfaces > Chat Widget in the sidebar.

What Is the Chat Widget?

The Chat Widget is a chat interface you can embed on any website or internal tool. It appears as either:

• A floating chat bubble in the corner of a webpage (like a support chat)
• An embedded chat panel built directly into a page

This is how you give your employees (or customers) access to the AI without asking them to log into Dokko’s admin dashboard.

Configuring the Chat Widget

The chat widget configuration has three tabs:

Tab 1: Chat Configuration

This controls the AI behavior:

Tab 2: Widget Configuration

This controls the appearance of the floating chat bubble:

Tab 3: Embedded Chat Configuration

For embedding a full chat panel inside an existing application (not a floating bubble), use this tab. Settings are similar but control the inline appearance.

Getting the Embed Code

Once you have configured the widget, you need to add it to your website. There are two options:

1. Download Script — downloads a JavaScript file. Upload this file to your website and include it with a <script> tag.
2. Generate Tag — generates an HTML snippet you can paste directly into your website’s HTML. Both options provide a small piece of code that, when added to any webpage, makes the chat widget appear.

Note: You will need basic web admin access (or help from your web developer) to add the code to your website.

5.2 Search Widget

Navigate to Interfaces > Search Widget in the sidebar.

The Search Widget is similar to the Chat Widget but presents a search box interface — more like a traditional search engine. Users type keywords and see a list of relevant document excerpts.

Configuration is similar to the Chat Widget. Connect it to your knowledge repositories, customize the appearance, and download the embed code to add it to your website.

Part 6: Agent Management — Automating Complex Tasks

Content Admins and Organization Admins: Agents are powerful automation tools. Set them up carefully.

6.1 Understanding Agents

Agent vs. Chat

Think of Chat as asking a question, and an Agent as delegating a project.

Agent Types

• LLM Agent: A single AI with a job description and a set of tools. The basic building block.
• Sequential Agent: Runs multiple sub-agents one after another, in a defined order. Like a pipeline: Agent A processes input, passes it to Agent B, which passes to Agent C.
• Parallel Agent: Runs multiple sub-agents at the same time. Useful when you need several independent pieces of information gathered simultaneously.
• Loop Agent: Repeats a sub-agent until a condition is met. Useful for iterative refinement or retry logic.

6.2 Creating an Agent

Navigate to Agent Management > Agent Configurations > Create.

Key Fields

Writing Good Instructions

Instructions are the most important part of an agent. Be specific:

• Vague instruction: “Help with HR questions.”
• Good instruction: “You are an HR policy expert for Acme Corp. When given a question, search the HR Policy repository, find the relevant policy, and provide a clear answer with the exact policy name and section number. Always conclude with: ‘For official guidance, please consult your HR Business Partner.’” After filling in the details, click Save. Test the agent using the Agent Run page (Part 4.2).

6.3 External Tool Servers (Adding Capabilities)

Navigate to Agent Management > Tool Servers in the sidebar.

What Is a Tool Server?

By default, an LLM Agent can search your documents and reason about them. But you can give agents additional abilities by connecting Tool Servers.

A Tool Server is an external service that exposes a set of actions the agent can call. Examples:

• A calendar service that the agent can check for meeting availability
• A database connector that the agent can query for customer data
• A web search tool that the agent can use to find current information

Tool Servers use a standard called MCP (Model Context Protocol) — think of it as a universal plug format for AI tools.

Adding a Tool Server

1. Click Add Tool Server or Create
2. Fill in:

• Name: a label for this tool server (e.g., “Company Database Connector”)
• Server URL: the web address of the tool server (provided by your IT team or the tool’s documentation)
• Authentication: the credentials needed to connect

3. Click Save Once saved, go back to an Agent Configuration and attach tools from this server to the agent’s available tools.

Note: Setting up tool servers requires technical knowledge. Involve your IT team if needed.

Part 7: Monitoring — Understanding Usage and Quality

Session Admins, Content Admins, and Organization Admins: This section covers monitoring conversations and improving quality.

7.1 Sessions

Navigate to Monitoring > Sessions in the sidebar.

What Is a Session?

A Session is one conversation thread — everything a user said and everything the AI responded, from the first message to the last. Every time someone opens the chat widget and starts typing, a new session begins.

The Sessions Page

The Sessions page shows two panels:

• Active Sessions: Conversations that are currently happening right now
• Intercepted Sessions: Conversations that an admin has joined to monitor For each session you can see: who started it (or if it was anonymous), when it started, how many messages have been exchanged, and the current status.

Session Interception

An admin can “intercept” an active session to monitor the conversation in real time — useful for quality assurance and support team supervision.

To intercept: click on a session and select Intercept. To release it: select Release.

7.2 Message Logs

Navigate to Monitoring > Message Logs in the sidebar.

Message Logs gives you a full searchable history of every message sent through your Dokko chat widgets.

Use it to:

• Review what questions users are asking most often
• Check if the AI gave a specific response correctly
• Audit usage for compliance purposes

You can filter logs by date range, session ID, or user.

7.3 Response Quality Analysis

Navigate to Monitoring > Initiate Analysis in the sidebar.

What Is Response Quality Analysis?

Dokko can run an automated quality scan over your past conversations. A second AI reads through the session history and flags responses that appear to be inaccurate, off-topic, or unhelpful.

Running an Analysis

1. Go to Monitoring > Initiate Analysis
2. Select:

• Document Source: which repository’s conversations to analyze
• LLM Configuration: which AI model to use for the analysis

3. Click Run Analysis
4. The analysis runs in the background. When complete, results appear in Response Overview

Reviewing Results

Go to Monitoring > Response Overview to see flagged responses. For each flagged item you can:

• Read the original question and the AI’s response
• Mark it as Accurate or Inaccurate
• Delete flagged items that are not relevant

Use this data to identify gaps in your knowledge base — if the AI often gives poor answers about a topic, upload better documents about that topic.

Part 8: Administration — Managing Your Organization

Organization Admins only: The Administration section is only visible to Organization Admins.

8.1 User Management

Navigate to Administration > Users in the sidebar.

Inviting New Users

1. Click Invite Users
2. Enter the user’s email address
3. Select their Role:

• Organization Admin: full control of everything
• Content Admin: manages documents and AI setup
• Session Admin: monitors sessions only
• User: can use chat only

4. Optionally assign Permissions (which document repositories they can access — see Part 8.2)
5. Click Send Invitation The user receives an email with a link to set up their account.

Managing Existing Users

On the Users page, you see a table of all users with their email, role, assigned permissions, and active/inactive status.

To deactivate a user (e.g., when an employee leaves): toggle the Active switch off. The user immediately loses access but their data is preserved.

To change a user’s role: find the user in the table, click the role dropdown, and select the new role.

Bulk actions: check the boxes next to multiple users to change their roles or permissions for all selected users at once — useful when onboarding a whole department.

8.2 Permissions

Navigate to Administration > Permissions in the sidebar.

What Is a Permission?

A Permission is like an employee access pass that grants the holder access to a specific set of document repositories.

Why you need permissions:

• Your sales team should only search the sales documents repository, not HR policies
• External users should only access public-facing knowledge, not internal reports
• Different departments have different document access needs

Creating a Permission

1. Click Create Permission
2. Fill in:

• Name: a descriptive label (e.g., “Sales Team Access”, “Executive Documents”, “Public Knowledge Base”)
• Repositories: select which repositories this permission grants access to

3. Click Save

Assigning Permissions to Users

1. Go to Administration > Users
2. Find the user and click Edit
3. In the Permissions section, check the boxes for the permissions to assign
4. Click Save Users can have multiple permissions. Their chat will search all repositories covered by their combined permissions.

8.3 Organization Settings

Navigate to Administration > Organization in the sidebar.

Here you can manage your organization’s basic settings:

• Organization Name: the name shown on the login screen and throughout the platform
• Allowed Email Domains: restrict who can self-register by listing your company’s email domains (e.g., yourcompany.com). Only people with matching email addresses can create accounts. After making changes, click Save.

8.4 Subscriptions

Navigate to Administration > Subscriptions in the sidebar.

Dokko has different subscription tiers that control which features are available to your organization.

The Subscriptions page shows your current plan and allows you to manage billing details.

Part 9: Troubleshooting & FAQ

“I can’t log in.”

Check these things in order:

1. Organization Name: this field is case-sensitive and must match exactly. Ask your Organization Admin to confirm the exact name.
2. Email address: make sure there are no typos and you are using the email the admin invited you with.
3. Password: click Forgot password? on the login page. A reset link will be sent to your email.
4. Account not activated: if you just received an invitation, make sure you clicked the link in the invitation email and set your password.
5. Account deactivated: contact your Organization Admin — they may have deactivated your account.

“The AI gives wrong or irrelevant answers.”

There are several possible causes:

1. Documents not uploaded: the relevant information has not been added to Dokko yet. Contact your Content Admin to upload the necessary files.
2. Ingestion not complete: documents take time to be processed and indexed. If documents were just uploaded, wait a few minutes and try again.
3. Permission issue: you may not have access to the repository containing the relevant documents. Ask your Organization Admin to check your permissions.
4. Wrong repository connected: the chat widget may not be connected to the correct repository. Ask your Content Admin to check the chat widget configuration.
5. AI temperature too high: if answers seem random or off-topic, the LLM configuration may have a high temperature. Ask your Content Admin to lower it toward 0.

“My uploaded document isn’t being found.”

1. Confirm the file was uploaded: go to Knowledge Base > File Management and look for the file.
2. Confirm the file is in a repository: a file must be in a folder that is marked as a Repository.
3. Check the ingestion status: navigate to the repository and look for the indexing status. It should say Complete. If it says Failed, there may be an issue with the file (corrupted, password-protected, unusual format).
4. Wait a few minutes: newly uploaded files can take a few minutes to be fully indexed.

“The crawler didn’t finish or failed.”

1. Open the starting URL in a regular browser and make sure it loads correctly.
2. Check if the site requires login: the crawler cannot access password-protected pages.
3. Reduce the maximum depth: very large sites with high depth settings can time out. Try depth 2 first.
4. Review any error messages shown in the crawler status.

“I don’t see a menu item that someone mentioned.”

This is almost always a role issue. Different roles see different menu items:

• Only My Chat visible → you are a User role
• Knowledge Base visible but not Administration → you are a Content Admin
• Everything visible → you are an Organization Admin

Contact your Organization Admin to check your role.

“How do I update documents that have changed?”

• For uploaded files: re-upload the new version to the same location in File Management.
• For crawled websites: navigate to the repository and trigger a new crawl. If auto-recrawl is configured, this happens on schedule automatically. After updating, wait a few minutes for the new version to be fully indexed before testing the chat.

“What file types does Dokko support?”

Dokko can process and index:

• PDF (.pdf) — most common
• Microsoft Word (.docx, .doc)
• Microsoft PowerPoint (.pptx, .ppt)
• Plain text (.txt)
• HTML (.html, .htm) — web pages
• Markdown (.md) — documentation files

For best results, use text-based PDFs (not scanned images). If you have scanned PDFs, enable the Foundation Model parsing strategy on your Document Source.

“Can people outside my company use the chat widget?”

Yes, with the anonymous chat feature. When “Require Login” is turned off in the Chat Widget configuration, anyone with access to your chat widget can use it without creating an account.

Note: Anonymous chat requires the Professional subscription plan. See Part 8.4.

“I see notifications in the sidebar — what do they mean?”

The bell icon in the sidebar shows system notifications for:

• Crawl job completed — a web crawl finished (successfully or with errors)
• Ingestion complete — documents have been indexed and are now searchable
• Ingestion failed — a problem occurred processing a document

Click the bell to see all notifications. You can mark them as read individually or all at once.

Appendix: Glossary

Agent An AI program configured to complete multi-step tasks autonomously. Unlike regular chat, an agent can use tools, search multiple sources, perform calculations, and execute a series of steps. Technical synonym: LLM Agent, AI Agent

API Key (Application Programming Interface Key) A private password-like string that grants Dokko permission to use an external AI service on your behalf. Usage is billed to the account that owns the key.

Chunking The process of splitting a large document into smaller pieces before indexing. Smaller pieces allow the AI to find the most relevant section of a document rather than returning the whole thing at once.

Configuration Set A bundle that groups multiple LLM configurations (chat, preprocessing) together under a single name. Applied to a document source or chat widget to ensure all parts of the AI pipeline use the right models.

Crawler / Web Crawler An automated program that visits web pages, reads their content, and imports it into Dokko’s knowledge base. You provide a starting URL and it follows links to gather content from the site.

Document Source The top-level cloud storage location for your organization’s documents. Connected to AWS S3 (Amazon cloud storage). Each organization has one, created automatically at registration.

Embedding A mathematical representation of text’s meaning — a list of numbers that captures semantic content so similar ideas can be found even when different words are used. Powers Dokko’s meaning-based search. Plain-English: a meaning fingerprint

Hybrid Search A search method that combines traditional keyword matching with meaning-based (embedding) search. Results are blended from both approaches, improving accuracy.

Ingestion Job The background process that takes an uploaded document, reads it, processes it, and adds it to the searchable index. Statuses: Starting, In Progress, Complete, Failed, Stopped

Knowledge Base The searchable index that Dokko builds from your documents. When a user asks a chat question, Dokko searches the knowledge base to find relevant information before generating an answer. Technical name: Bedrock Knowledge Base (AWS)

LLM (Large Language Model) The AI model that reads questions and writes answers. Examples: GPT-4o (OpenAI), Claude (Anthropic), Gemini (Google). You choose which LLM to use via LLM Configurations. Plain-English: the AI brain

MCP Server (Model Context Protocol Server) A tool server that gives an AI agent new capabilities — like accessing a database, reading a calendar, or searching the web. MCP is the standard format for connecting external tools to AI agents.

Permission A named access pass that grants a user access to specific repositories. Admins assign permissions to control which document collections each user can search.

Prompt Template A background instruction given to the AI that shapes its behavior for every conversation — persona, response style, topic restrictions, or language. Users never see this instruction; they only experience its effect.

RAG (Retrieval-Augmented Generation) The core technique Dokko uses: before the AI generates an answer, it retrieves relevant passages from your documents, then uses those passages as the factual basis for its generated response. Plain-English: look it up, then answer

Repository A named collection of documents within a document source. Repositories organize your knowledge into logical groups. Permissions and chat widgets connect to specific repositories.

Reranking A second-pass quality check: after initial retrieval finds candidate document chunks, a specialized AI model re-evaluates and re-scores them by relevance. Improves answer accuracy.

Session One conversation thread between a user and the chat interface — all messages exchanged from when the chat was opened until closed or a new conversation started.

Tenant Your organization’s private, isolated workspace within Dokko. All users, documents, configurations, and data belong to a specific tenant. Tenants are completely separate from each other. Plain-English: your organization’s Dokko workspace

Temperature A setting that controls how creative or random the AI’s responses are. 0 = very precise and predictable; 2 = very creative and variable. For factual knowledge-base chat, use 0.0–0.3.

Token The unit of text that AI models process. Roughly, 1 token ≈ ¾ of a word. “Max Tokens” controls the maximum length of each AI response. 1,000 tokens ≈ 750 words.

Widget An embeddable interface — a self-contained chat or search box you add to any website by inserting a snippet of JavaScript code. Dokko offers a Chat Widget and a Search Widget.

End of Dokko User Manual

Questions or feedback? Contact your Organization Admin, or reach out to your Dokko support contact.