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 a 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 the content you’ve added and writes an answer using it 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 connected 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.

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 add your files — PDFs, Word documents, web pages — and Dokko processes them so their meaning can be searched.

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 searches your documents, and the AI writes an answer using those documents as its source. At the bottom of an answer you can see Sources — links to the exact documents used.

The key thing to understand: Dokko’s AI answers from what you give it. A document-based chat does not browse the open internet — it answers based on the documents you’ve added. 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 also see sections like “Knowledge Base”, “Agent Management”, 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. Create a repository, then add documents or crawl a website (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. Create a repository (Part 2.2)
3. Upload documents or crawl a website into it (Part 2.3)
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 contains a button or 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 opens a web page where you set your password.

1.2 Setting Your Password

On the registration page you will see:

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

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

Your email address is shown but cannot be changed — it is fixed to the address you were invited with.

Click Save. Your account is now active and you are logged in.

1.3 Logging In

Go to your organization’s Dokko web address in a browser. The login page has three fields:

1. Organization Name — the name of your company’s Dokko workspace (written in your invitation email)
2. Email — your work email address
3. Password — the password you created

Click Log In.

Forgot your password? Click the Forgot password? link. Enter your email address and your organization name, and you will receive a password reset email.

1.4 The Dashboard — What You See First

After logging in, Organization Admins and Content Admins land on the Dashboard — an overview page summarizing activity across your organization, such as active sessions, configured agents, and the status of your AI providers.

If you are a Session Admin, you land on the monitoring area. If you are a regular User, you land directly on My Chat.

1.5 The Sidebar Navigation

On the left side of the screen is the sidebar — the main navigation menu, 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 account menu and a Logout option.

1.6 Account Settings

To manage your personal account, open Account Settings from your account menu. There you can see your email and role, and:

• Change your password — enter your current password, then your new password twice, and click Save.
• Delete your account — click Delete Account. For safety, you must type the first part of your email address (the portion before the @) to confirm.

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 manage the Knowledge Base.

2.1 How Your Knowledge Is Organized

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

Knowledge Base   (Everything Dokko has indexed for your organization)
└── Repository   (A named collection, e.g., "Legal", "HR", "Product Docs")
    └── Files     (Individual documents, or pages imported from a website)

• Knowledge Base: Your organization’s overall searchable library. It is set up automatically for your organization, and you tune how it processes and searches documents from one Configuration page (Part 2.4).

• Repository: A named collection of documents. You can have many repositories — one for “HR Policies”, one for “Product Manuals”, one for “Customer Contracts”, and so on. Repositories are the unit you organize and control access to.

• Files: The actual documents inside a repository — uploaded files, or pages imported from a website by the crawler.

Why does this structure matter? Access to documents is controlled per repository through Permissions (Part 8.2). This means your sales team only searches sales documents, HR only sees HR policies, and so on. Good organization leads to better, more focused answers.

The correct order is: create a repository first, then add documents to it. (Earlier versions let you “mark a folder as a repository” — that no longer exists.)

2.2 Repositories — Creating a Collection

Navigate to Knowledge Base > Repositories in the sidebar.

A Repository is a named collection of documents the AI can search. This is your starting point — you create the repository first, then fill it with content.

Creating a Repository

1. Click Create (or the + button)
2. Fill in:
   • Name: a clear, descriptive name (e.g., “HR Policies”, “Product Documentation”, “Customer FAQs”)
   • Type: choose how this collection will be used:
     • Semantic — the standard choice. Documents are indexed so the AI can answer questions about them in chat.
     • Web Search — a collection geared toward keyword-style search results rather than conversational answers.
   • Permissions: select which permissions grant access to this repository. Anyone (a user or a chat widget) holding one of these permissions will be able to search it. See Part 8.2 to create permissions first if you don’t have any yet.
3. Click Save

Once the repository exists, add documents to it through File Management — either by uploading files or by crawling a website (see below).

2.3 File Management — Adding Documents and Crawling Websites

Navigate to Knowledge Base > File Management in the sidebar.

File Management is the document browser for a repository. It always works inside one repository at a time — first select a repository, then you see and manage its files. (If you have no repositories yet, create one first in Part 2.2.)

Uploading Files

1. Select the repository you want to work in
2. Navigate to the folder where you want to upload (or stay at the root)
3. Either:
   • Drag and drop files from your computer onto the file browser, or
   • Use the Upload action to browse and select files
4. You can upload multiple files at once; a progress indicator shows how the upload is going

Supported File Types

Dokko can read and index PDFs, Microsoft Word documents, PowerPoint files, plain text, HTML pages, Markdown, and more. For best results with PDFs, use text-based files rather than scanned images (see Part 2.4 for handling complex or scanned documents).

Organizing with Folders

Use the right-click menu (or the toolbar) to create folders, and to rename, move, or delete files and folders. You can also edit a file’s metadata, and edit the contents of text-based files in place.

Warning: Deleting a file removes it from the repository and from search. This cannot be undone. Make sure you have the original elsewhere before deleting.

Crawling a Website

If your knowledge lives on a website or intranet, Dokko can read and import those pages for you. This is called web crawling, and you start it from File Management inside the target repository.

1. With a repository selected, click Crawl
2. Fill in the crawl settings:
   • Starting URL: the web address where the crawler begins (e.g., https://help.yourcompany.com)
   • Crawl Type: the kind of crawl to run (a standard site crawl is the usual choice)
   • Maximum Depth: how many clicks deep to follow links — depth 1 is the starting page only, depth 2 also follows its links, and so on. Start with 2 or 3; too deep can import thousands of irrelevant pages. Set it to -1 for no limit.
   • Included URLs (optional): only crawl pages whose addresses contain these patterns (e.g., /docs/)
   • Excluded URLs (optional): skip pages matching these patterns (e.g., /blog/)
3. Click to start the crawl

The crawler runs in the background, and the repository is temporarily locked while it works. When it finishes, the new pages appear in the repository and are indexed for search.

Scheduled Re-Crawling

If the website changes regularly, you can have Dokko re-crawl it automatically. In the crawl settings:

• Recrawl Starting Date: when the first automatic recrawl should run
• Recrawl Frequency (days): how often to repeat it

You can turn an existing schedule off again with Disable Auto Recrawl. There’s also a Force Recrawl option to re-import pages even if they were already indexed.

This keeps your knowledge base up to date without manual work.

2.4 Knowledge Base Configuration — Tuning Quality

Navigate to Knowledge Base > Configuration in the sidebar.

This single page is where you tune how Dokko processes your documents and how it searches them when answering. These settings apply across your knowledge base. For most organizations the defaults work well — adjust them only when you want to improve results for a particular kind of content.

The page shows a status indicator at the top:

How Documents Are Read

Parsing Strategy — how Dokko “reads” your documents:

How Documents Are Split

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

Contextual Expansion — when enabled, each piece is kept aware of the broader document it came from, so its meaning stays clear even when read on its own. This noticeably improves results, especially for short or ambiguous passages, at a small one-time processing cost. When Contextual Expansion is on, chunking is handled for you (set to None).

How Answers Are Retrieved

These control what Dokko pulls from your documents before writing an answer:

• Hybrid Search — searches two ways at once: plain keyword matching and meaning-based search (finding relevant passages even when different words are used). Recommended for most cases.
• Number of results to retrieve — how many passages Dokko gathers as candidate material for an answer. More can mean richer answers but a bit more processing and token consumption. The usual range is a handful of results; it can go higher when reranking is on.
• Reranking — after the first set of candidates is gathered, a second pass re-scores them for relevance so the best material rises to the top. You can turn it on, and set how many top results to keep. Improves accuracy at a small cost in processing time.

Helping the AI Filter by Metadata

• Implicit Filtering — lets the AI narrow results using your documents’ metadata automatically, based on what the question implies (for example, focusing on a particular document type or category).
• Metadata Schema — define the custom metadata fields your documents carry (a name, a type such as text/number/true-false/list, a short description, and a hint about how it should be used for filtering). This gives the filtering above something concrete to work with.

Note: To change parsing, chunking strategy and contexutal enrichement, you have to re-initalize the Knowledge base (destroy and create new).

Part 3: Infrastructure — Setting Up the AI Engine

Organization Admins: You set up and change the AI engine here. Content Admins: You can view these settings but cannot change them. 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 from a provider like OpenAI or Anthropic, Dokko needs permission to use that service on your behalf. That permission comes as an API key — essentially a private password tied to an account with the AI provider.

You don’t have to bring your own keys. Dokko provides built-in default keys for each supported provider, so you can start using the AI right away without signing up with the providers yourself. When you use the default keys, Dokko tracks your usage per provider and bills it according to your subscription tier’s rules. You can still add your own API keys if you’d rather have provider usage billed directly to your own provider account — in that case the provider bills you directly, based on whose key is used.

Supported AI Providers

The exact models available depend on the provider you choose and update over time.

How to Get an API Key

1. Go to the provider’s website and create an account
2. Open their API Keys section (usually under Settings or Account)
3. Create a new key
4. Copy the key immediately — most providers only show it once

Adding an API Key to Dokko

1. Click Create (or Add Key)
2. Fill in:
   • Name: a label so you remember what this key is for (e.g., “OpenAI Production Key”)
   • Provider: select the AI provider
   • Key: paste the key you copied
3. Click Save

Security note: Dokko stores your API key securely. 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 is a recipe that tells Dokko which AI model to use, how it should behave, and how its background instructions read. You can create several configurations for different purposes — one for the public chat widget, one for internal use, one for analysis etc.

Creating an LLM Configuration

1. Click Create
2. Fill in the fields:

3. Click Save

Tip on Temperature: For a company knowledge-base assistant that should give accurate, grounded answers, use a low temperature (around 0.0–0.3). Higher values suit brainstorming or creative writing.

The Prompt Template Field — Shaping the AI’s Behavior

The Prompt Template is a background instruction you give the AI, written right here in the configuration. The user never sees it, but it shapes every response. Use it to:

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

Leave it blank if you don’t need any special behavior.

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 field at the bottom
2. Type your question in plain language — exactly as you’d ask a colleague:
   • “What is our policy on remote work?”
   • “Who is the contact person for IT support?”
   • “Summarize the key points from the Q3 report.”
3. Press Enter or click Send
4. The answer appears, streamed word by word in real time

Reading the Response

At the bottom of an answer you may see a Sources section — the exact documents the AI used. Click a source to open the original document.

Important: If the AI says it doesn’t have information about something, the relevant documents either haven’t been added yet, or you don’t have permission to access the repository that contains them. Contact your Content Admin.

Continuing a Conversation

Your conversation is saved as a session. Keep typing follow-up questions in the same session and the AI remembers the context — if you asked about “the leave policy” earlier, you can simply 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 — searching multiple sources, performing calculations, looking up external systems, and combining results into a structured response. Think of it as the difference between asking a question and delegating a task.

Running an Agent

1. On the Agent Run page, select an agent (its details, such as model and instructions, are shown for reference)
2. Type your request in the input field
3. Send it
4. Watch the agent work in real time:
   • You’ll see function calls appear — the steps the agent is taking (e.g., searching documents, running a web lookup, running a calculation)
   • Each step shows its result
   • Finally, the agent writes a complete response

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

Content Admins and Organization Admins: This section covers deploying Dokko’s chat 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. When you create one, you choose its mode:

• Widget — a floating chat bubble in the corner of a page (like a support chat)
• Embedded — a chat panel built directly into a page

This is how you give employees or customers access to the AI without asking them to log into the Dokko dashboard.

Configuring the Chat Widget

A chat widget configuration has two tabs:

Tab 1: General

This controls behavior and content:

You can use Test Config to try the chat live while you set it up.

Tab 2: Appearance

This controls the look of the widget — colors for the header, message bubbles, input area, and launcher button, with optional dark mode. There’s a handy Generate from Brand Color tool: pick your brand color and Dokko proposes a coordinated palette. A Reset Defaults option restores the standard styling.

Getting the Embed Code

Once configured, open the Embed Code option to get the code for your website. The dialog offers snippets for plain JavaScript, React, Vue, and Angular — pick your framework, copy the snippet, and add it to your site. For an embedded chat, you also place a small container element where the chat should appear. Copy buttons are provided for each snippet.

Note: You’ll 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 will provide a search-box interface — more like a traditional search engine — where users type keywords and see a list of relevant document excerpts with relevance scores.

Coming soon: The Search Widget is not yet available to configure. This page currently shows a “Coming Soon” notice. This manual will be updated when it ships.

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 (its instructions), a chosen model, and an optional set of tools. The basic building block.
• Sequential Agent: Runs several sub-agents one after another, in order — like a pipeline.
• Parallel Agent: Runs several sub-agents at the same time, useful when you need independent pieces of information gathered simultaneously.
• Loop Agent: Repeats a sub-agent until a condition is met or a maximum number of iterations is reached — useful for iterative refinement.

Sequential, Parallel, and Loop agents are built by combining other agents (their sub-agents); LLM agents are where the actual instructions, model, and tools live.

6.2 Creating an Agent

Navigate to Agent Management > Agent Configurations, then create a new agent.

Key Fields

LLM agents also offer advanced options (such as additional global instructions and structured input/output) for power users; you can ignore these for a basic agent.

Writing Good Instructions

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

• Vague: “Help with HR questions.”
• Good: “You are an HR policy expert. 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. Always conclude by advising the reader to confirm with their HR contact for official guidance.”

After saving, test the agent on the Agent Run page (Part 4.2).

6.3 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. You can give agents additional abilities by connecting Tool Servers — external services that expose actions an agent can call, such as:

• A database connector the agent can query
• A web search tool the agent can use to find current information
• A calendar or other business system the agent can check

Adding a Tool Server

1. Click Create
2. Fill in:
   • Name: a label for this server
   • URL: the address of the tool server
   • Type: the kind of server — MCP (Model Context Protocol) or OpenAPI
   • Credentials (MCP): the authentication method (API key or bearer token) and the secret value
   • Config (OpenAPI): the server’s configuration details
3. Click Save

Attaching Tools to an Agent

Open an LLM agent’s configuration and use its Tool Servers tab to attach a server and pick which of its tools the agent may use. (Only LLM agents can use tool servers.)

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

Part 7: Monitoring — Understanding Usage and Quality

Session Admins can use Sessions and Message Logs. Content Admins and Organization Admins can also use Response Analytics.

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. A new session begins each time someone opens a chat and starts typing.

The Sessions Page

The page has two tabs:

• Active Sessions — conversations happening right now
• Intercepted Sessions — conversations an admin has joined to monitor

For each session you can see who started it (or whether it was anonymous), which widget it came from, the last activity time, and which admin (if any) is handling it.

Joining a Live Session

From the Active Sessions tab, click Take Over on a session to open it. You first see it in a read-only preview; once you take over, you can monitor it in real time and send live responses if needed. When you’re done, click Release (or Release All to release every session you’re monitoring at once).

7.2 Message Logs

Navigate to Monitoring > Message Logs in the sidebar.

Message Logs is a full history of messages exchanged through your chat widgets. For each entry you can see the widget, the user (or “Anonymous”), the user’s message, the assistant’s response, the date, and how many document passages were used.

Click any entry to open a detail panel showing the full exchange and the exact passages (with source links) the AI drew on. Browse through the history page by page.

Use it to review what users are asking, confirm specific answers, and audit usage.

7.3 Response Analytics

Navigate to Monitoring > Response Analytics in the sidebar.

What Is Response Analytics?

Dokko can run an automated quality scan over your past conversations. A separate AI reads through them and judges whether each response looks factually sound, flagging the weak ones along with the material that produced them. You run the scan and review the results on this one page.

Running an Analysis

1. Click Run Analysis
2. Choose:
   • LLM Configuration: which configuration to use for the scan (this must be a configuration of type Bad Response Analysis — see Part 3.2)
   • How many recent messages to analyze
   • Optionally, a random sample size to spot-check rather than scan everything
3. Start it. The scan runs in the background; if one is already running, Dokko tells you to wait for it to finish.

Reviewing Results

Results appear on the same page, grouped into tabs:

• All — every analyzed response
• Good — judged factually sound
• Bad — flagged as questionable

For each item you can read the original question, the AI’s response, and the reason the analysis gave for its judgment. Click an item to open a detail panel with the full exchange and the document passages used. If a flagged item isn’t useful, you can exclude it to remove it from the list.

Use this to find gaps in your knowledge base — if the AI repeatedly answers a topic poorly, add or improve documents on that topic.

Part 8: Administration — Managing Your Organization

Organization Admins manage Users and Organization settings. Content Admins can also manage Permissions.

8.1 User Management

Navigate to Administration > Users in the sidebar. (Organization Admins only.)

Inviting New Users

1. Click Invite Users
2. Enter one or more email addresses (you can invite several at once)
3. Select a Role:
   • Organization Admin: full control
   • Content Admin: manages documents, agents, interfaces, permissions, and monitoring
   • Session Admin: monitors sessions and message logs
   • User: can use chat only
4. Optionally assign Permissions (which repositories they can access — see Part 8.2)
5. Send the invitations

Each user receives an email with a link to set up their account.

Managing Existing Users

The Users page lists everyone with their email, role, permissions, verification, and active status. You can:

• Change a user’s role — individually, or for several users at once using the bulk action
• Activate or deactivate a user — deactivating immediately removes access while preserving the user’s data (useful when someone leaves)
• Adjust a user’s permissions
• Delete a user

8.2 Permissions

Navigate to Administration > Permissions in the sidebar. (Organization Admins and Content Admins.)

What Is a Permission?

A Permission is a named access pass. It connects two things:

• Repositories declare which permissions grant access to them (set when you create or edit a repository — Part 2.2).
• Users and chat widgets are granted permissions.

Anyone (or any widget) holding a permission can search every repository that lists that permission. This is how you keep the sales team in sales documents, HR in HR policies, and external users limited to public content.

Creating a Permission

1. Click Create
2. Give it a clear Name (e.g., “Sales Team Access”, “Executive Documents”, “Public Knowledge Base”)
3. Click Save

Then link it to repositories (in the repository’s settings) and assign it to users (in User Management) or to a chat widget (in the widget’s settings). Users can hold several permissions and will search the combined set of repositories.

8.3 Organization Settings

Navigate to Administration > Organization in the sidebar. (Organization Admins only.)

Here you manage your organization’s basic settings:

• Organization Name: the name shown on the login screen and throughout the platform
• Allowed Domains: restrict where your chat widget can be placed. Before integrating the chat widget on a site (for exmaple https://dokko.ai) you need to allow this domain.

After making changes, click Save.

Part 9: Troubleshooting & FAQ

“I can’t log in.”

Check these in order:

1. Organization Name: case-sensitive and must match exactly. Ask your Organization Admin to confirm it.
2. Email address: no typos, and the same address you were invited with.
3. Password: use Forgot password? to reset it (you’ll need your email and organization name).
4. Account not activated: if you just received an invitation, make sure you clicked the link and set your password.
5. Account deactivated: contact your Organization Admin.

“The AI gives wrong or irrelevant answers.”

1. Documents not added: the relevant information isn’t in Dokko yet. Ask your Content Admin to add it.
2. Still processing: newly added documents take a little time to be indexed. Wait a few minutes and try again.
3. Permission issue: you may not have access to the repository with the relevant documents. Ask an admin to check your permissions.
4. Wrong access connected: the chat widget may not hold the permission for the right repository. Ask your Content Admin to check the widget’s settings.
5. Temperature too high: if answers seem random or off-topic, the LLM configuration’s temperature may be too high. Ask your Organization Admin to lower it toward 0.

“My uploaded document isn’t being found.”

1. Confirm the file is inside a repository: go to Knowledge Base > File Management, select the repository, and look for the file.
2. Check the status on the Configuration page: it should reach Ready. If processing failed, the file may be corrupted, password-protected, or in an unusual format.
3. Wait a few minutes: newly added files take time to be fully indexed.

“The crawler didn’t finish or failed.”

1. Open the starting URL in a regular browser and confirm it loads.
2. Check whether the site requires login — the crawler cannot access password-protected pages.
3. Reduce the maximum depth: very large sites with high depth can time out. Try depth 2 first.
4. Remember the repository is locked while a crawl runs; wait for it to finish before starting another.

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

This is almost always a role difference:

• Only My Chat visible → you are a User
• Knowledge Base / Agent Management visible but not Administration → you are likely a Content Admin
• Only monitoring visible → you are a Session Admin
• Everything visible → you are an Organization Admin

Note that some Infrastructure settings are view-only for Content Admins and editable only by Organization Admins. Contact your Organization Admin to check your role.

“How do I update documents that have changed?”

• For uploaded files: upload the new version to the same repository location in File Management.
• For crawled websites: run the crawl again (or rely on a recrawl schedule if one is set).

After updating, wait a few minutes for the new version to be indexed before testing the chat.

“What file types does Dokko support?”

PDF, Microsoft Word, PowerPoint, plain text, HTML pages, and Markdown, among others. For best results use text-based PDFs rather than scanned images. For scanned or visually complex documents, switch the Parsing Strategy to Foundation Model on the Knowledge Base Configuration page (Part 2.4).

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

Yes. Turn Require Login off in the chat widget’s settings, and anyone with access to the widget can use it without creating an account.

“I see notifications — what do they mean?”

Notifications keep you informed about background activity, such as a web crawl finishing or documents finishing (or failing) processing. Open the notifications area to review them and mark them as read.

Appendix: Glossary

Agent An AI configured to complete multi-step tasks. Unlike regular chat, an agent can use tools, search multiple sources, and execute a series of steps.

API 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 Splitting a large document into smaller pieces before indexing, so the AI can find the most relevant section rather than returning a whole document.

Crawler / Web Crawler An automated process that visits web pages, reads their content, and imports it into a repository. You provide a starting URL and it follows links to gather content.

Hybrid Search A search approach that combines keyword matching with meaning-based search, blending both for better accuracy.

Knowledge Base Your organization’s overall searchable library, built from the documents you add. When a user asks a question, Dokko searches the knowledge base for relevant passages before answering. It is set up automatically and tuned from the Configuration page.

LLM Configuration A saved recipe that selects an AI model, sets how it behaves (temperature, response length), and holds optional background instructions (the Prompt Template).

Permission A named access pass. Repositories declare which permissions can access them; users and chat widgets are granted permissions. Holding a permission grants access to every repository that lists it.

Prompt Template Background instructions, set inside an LLM Configuration or an agent, that shape the AI’s behavior — persona, style, topic limits, or language. Users never see it; they only experience its effect.

RAG (Retrieval-Augmented Generation) The approach behind Dokko’s grounded answers: it first retrieves relevant passages from your documents, then writes an answer based on them. In plain English: look it up, then answer.

Reranking A second-pass quality check: after the first set of candidate passages is gathered, a model re-scores them by relevance so the best material is used. Improves answer accuracy.

Repository A named collection of documents. Repositories organize your knowledge and are the unit that access (permissions) is controlled on. Each repository has a type — Semantic (for chat/Q&A) or Web Search.

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

Temperature A setting that controls how creative or predictable the AI’s responses are. 0 = very precise; higher = more varied. For factual knowledge-base chat, use a low value (0.0–0.3).

Token The unit of text AI models process. Roughly, 1 token ≈ ¾ of a word. “Max Tokens” caps the length of a response.

Tool Server An external service (MCP or OpenAPI) that gives an agent new abilities — such as querying a database, searching the web, or checking a business system.

Widget An embeddable interface you add to a website with a small code snippet. Dokko offers a Chat Widget (and a Search Widget, coming soon).

End of Dokko User Manual

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