Before your Korrah Assistant can start learning, she needs a dedicated workspace — a Project. Projects keep your assistants organised and ensure that each one draws knowledge from the right set of materials.
From the left-hand menu, go to Projects and select Add New.
Give your project a clear, descriptive name — for example, “Client Onboarding FAQs” or “TaxNav Help Centre.”
Every project starts with at least one Data Collector, which defines where Korrah learns from. You can choose between:
If you already have structured documents — such as training manuals, policy guides, or Q&A documents — you can upload them directly.
Each uploaded document automatically generates up to 10 questions, so try to keep your PDFs concise and focused on a single topic. Shorter, topic-specific files yield much higher-quality answers.
Alternatively, you can use a website crawler to import knowledge directly from your own or public websites. The crawler allows Korrah to read and understand your content just like a human researcher would.
You have two ways to define how the crawler works:
1. Pattern Mode
Provide a main URL and specify the depth of the crawl (how many levels deep the crawler should go).
This option is ideal for:
Use the whitelist and blacklist settings to include or exclude specific pages. This helps prevent irrelevant links — such as Terms & Conditions or unrelated content — from being scraped.
2. URL List Mode
Instead of crawling automatically, you can manually upload a list of URLs you want Korrah to learn from.
This approach works best for:
If your use case doesn’t fit neatly into one of these scenarios, don’t worry — our team can help you design the most effective setup for your data.
Once your content is uploaded, Korrah will begin processing it in the background — analysing, indexing, and preparing it for use. This may take a little time, depending on the size of your data sources.
While the system works its magic, you can customise your assistant’s look and feel — choose her name, avatar, tone of voice, and chat interface settings to match your brand identity. By the time the data is ready, your Korrah will not only know what to say but also look and sound exactly the way you want her to.
Once your project and data collectors are ready, it’s time to create your Assistants — the personalities and interfaces that interact directly with your users.
You can create multiple assistants that draw from the same data sources. This flexibility is useful if, for example, you want one public-facing assistant for your website and another private assistant for internal use. Both can share the same knowledge base but differ in tone of voice, permissions, and behaviour.
The System Prompt is the heart of your assistant’s personality. It defines how Korrah behaves, what tone she uses, and how she responds in specific situations.
Think of it as writing an instruction note to a new team member. For example:
You can write your own prompt from scratch or use the auto-generate feature as a starting point, then refine it to match your brand and communication style.
This section allows you to enable additional capabilities for your assistant.
Enable Assistant Widget
This option activates the widget — the chat avatar that can be embedded on your website.
Enable Tools, Tasks, and Article Categories
These are agentic AI features that extend Korrah’s capabilities — enabling her to take actions, manage workflows, and categorise content intelligently.
If you’d like to activate these features, please contact us to discuss your use case and configuration options.
This is where Korrah comes to life. In the Widget Preview, you can customise her appearance and personality:
By tailoring these details, you ensure Korrah not only speaks accurately — but also reflects your organisation’s voice, tone, and brand identity.
With your project and assistants set up, it’s time to focus on what really powers Korrah — her Knowledge Base. This is where all the information she uses to answer questions is stored, refined, and tested.
Your Knowledge Base is the collection of documents and web pages ingested by your Data Collectors. Think of the Data Collector as the workhorse — it fetches and organises information — while the Knowledge Base is its output, containing the Q&A pairs that Korrah learns from.
From the Knowledge Base view, you can explore each imported component, review the automatically generated Q&A pairs, and even edit or add new ones manually.
Sometimes, you’ll find that certain questions don’t have answers within your existing data sources. To address this, you can add Q&A pairs manually — but there’s a small trick to doing it right.
Since every Q&A must belong to a data source, start by creating a “technical” data source — for example, a simple PDF containing a handful of placeholder questions. You can then add your custom Q&A pairs to that file. This is a great way to handle exceptions or special topics that don’t neatly fit within your main documents.
Once your Knowledge Base feels complete, it’s time to test how well Korrah performs.
Go to the Playground — your private sandbox environment — and start chatting with your assistant. Ask real-world questions and pay attention to the quality of her answers. If the responses are accurate and natural, congratulations — you’re on the right track. If not, consider the following:
You can then:
Repeat this process until you’re satisfied with the quality and consistency of the answers.
At this stage, your Queries tab will still be empty — it only begins to populate once your assistant is published and live users start interacting with her. That’s the second phase of testing.
For now, focus on perfecting Korrah in the sandbox. Once you’re happy with her performance, launch the bot and invite your team or a small group of trusted customers to test it in real-world conditions. Their feedback will help you fine-tune the assistant before a full public release.
Congratulations — you’ve launched your assistant!
But the journey doesn’t end here. Before Korrah can truly be considered battle-tested, she needs to be refined based on real-world interactions. This final stage is about listening, learning, and fine-tuning to ensure your assistant consistently delivers high-quality answers.
Once your assistant is live, the Queries tab becomes your central dashboard for improvement.
Here, you’ll see a record of all user interactions — the questions asked, Korrah’s answers, and an accuracy score indicating how confident she was in each response.
Review these regularly and take corrective action where needed:
Small, continuous improvements at this stage can dramatically increase both the accuracy and personality of your assistant.
Korrah’s learning environment doesn’t stand still — and neither should your oversight.
If your assistant draws from dynamic data sources such as live websites that change content frequently, you’ll need to adjust the re-scraping frequency.
Keep in mind that each re-scrape consumes processing tokens, but this is a necessary investment when working with constantly evolving information.
Set a regular review cycle — weekly or monthly — to keep your assistant aligned with your most up-to-date knowledge and brand tone.
Once you’ve mastered these steps and your assistant runs smoothly, you’ll be ready to explore the next frontier — Agentic AI Assistants. These advanced versions of Korrah can take actions, perform tasks, and manage workflows autonomously.
We’ll cover that exciting topic in a dedicated blog post soon — so stay tuned.
