Skip to main content
QA.tech’s knowledge system helps AI assistants understand your project and helps the test agent run reliably. Assistant knowledge shapes how the chat assistant answers questions and creates tests, while Test Agent knowledge guides the agent as it executes those tests. Together they bring more accurate test generation and better project-specific guidance.

Settings → Knowledge

Knowledge lives in a sidebar group under Settings → Knowledge, split into three tabs. Opening Settings → Knowledge lands you on the Assistant tab.

Assistant tab

Knowledge used by the AI Chat Assistant and during test creation.

Project Summary

An AI-generated summary of your website’s functionality and structure, created by analyzing your pages and user interactions. The summary is always part of chat context and updates as your site evolves. What’s Included:
  • Overview of your platform’s purpose and main functionality
  • Key features and user flows
  • Important pages and their purposes
  • Integration points and external services
To Refresh the Summary:
1

Navigate to Knowledge Settings

2

Find Project Summary

Find the “Project Summary” section on the Assistant tab.
3

Refresh Summary

Click “Refresh” to generate an updated summary based on your latest site analysis.
The summary requires initial site analysis to be completed. If no summary appears, ensure you’ve run some tests or site analysis first.

Assistant Rules

Rules the chat assistant always follows when answering questions and creating tests. See Rules below.

PR Review Rules

Rules applied during PR review. See Rules below.

Library

Link and text knowledge items that AI assistants can reference when generating tests and answering questions. Library items are stored at the project level and shared across all team members. What You Can Add:
  • URLs: Link to documentation sites, help pages, API docs. URLs are crawled and indexed.
  • Text Content: Add custom instructions, project notes, or guidelines. Text items are stored and summarized.
See Adding Library Items for steps.

Memories

Knowledge items the assistant creates automatically from your conversations, capturing useful facts it learns as you work. You can review, edit, and delete memories from the Assistant tab so the assistant keeps an accurate picture of your project.

Test Agent tab

Knowledge used by the test agent during test execution.

Agent Rules

Rules the test agent follows while executing tests. Agent Rules support targeting filters - by application, labels, scenario, or URL path - so a rule applies only to matching tests instead of every test in the project. See Rules below.

Agent Visuals

Image uploads that help the agent visually identify UI elements during execution. Add screenshots of buttons, icons, or layouts the agent should recognize so it can act on them reliably.

Crawling tab

Manage crawling sessions, your project domains (suggested domains and excluded paths), and the Reset Knowledge Graph action. See Crawling Sessions for how crawling builds the agent’s understanding of your application.

Rules

Rules are how you give the AI persistent instructions that should always apply. There are three kinds, each used at a different stage:
  • Assistant Rules - followed by the chat assistant during chat and test creation.
  • PR Review Rules - applied during PR review.
  • Agent Rules - followed by the test agent during test execution. Agent Rules support targeting filters (by application, labels, scenario, or URL path) so a rule applies only to matching tests.
As a guideline: execution-relevant guidance belongs in Agent Rules, while creation-relevant guidance belongs in Assistant Rules. Good Rules - Use Cases with Examples:
Use CaseExampleWhat It Achieves
Domain knowledge”This is a railway booking system for American routes. Stations use AMTRAK codes.”Agent understands domain terminology
Quality standards”Tests should fail if obvious typos or broken layouts are detected.”Agent enforces quality expectations
Data requirements”When creating test users, always use Swedish names and addresses.”Agent generates appropriate test data
UI conventions”Red buttons indicate destructive actions - verify confirmation dialogs appear before clicking.”Agent handles UI patterns correctly
Auth flows”Login requires SMS 2FA. The code will be available in the test inbox.”Agent knows to check inbox for 2FA
Timing expectations”Page loads may take up to 10 seconds in staging. Wait for loading indicators to disappear.”Agent waits appropriately
Known limitations”The checkout flow is broken on mobile viewports - skip mobile tests for checkout.”Agent avoids known issues
Best Practices:
  • Keep rules concise but specific - Agent Rules are consulted on every relevant action.
  • Focus on guidance that applies broadly; use Agent Rule targeting filters when something should apply only to certain tests.
  • Include domain-specific terminology and behaviors.
  • Mention any unusual UI patterns or interactions.
  • Update rules when your product behavior changes.
What NOT to Include:
  • Test-specific instructions - put these in test steps instead
  • Credentials - use Configs instead
  • URLs - use Applications and Environments instead
  • Temporary workarounds - document in test steps where needed

How Knowledge Works

Knowledge has different roles during test creation versus test execution:
Knowledge TypeTest Creation (Chat)Test Execution
Library (link/text items)Used - searched semanticallyNot used (unless attached to the test)
MemoriesUsedNot used (unless attached to the test)
Knowledge GraphUsed - site structure + semantic searchNot used
Project SummaryUsedUsed (via agent context)
Assistant RulesUsedNot used
Agent RulesNot usedUsed - every relevant decision (subject to targeting filters)
Agent VisualsNot usedUsed
Chat file uploads (attachments)Used in that chat onlyNot used
During Test Creation (Chat & Test Generation):
  • The assistant searches your knowledge on demand - semantic search across your Library, Memories, and the site/feature knowledge graph - rather than stuffing everything into the prompt.
  • The Project Summary provides overall context about your application.
  • Assistant Rules shape how the assistant answers and what tests it proposes.
  • The AI generates test steps based on all the above context.
During Test Execution: Only a targeted subset of knowledge is available to the test agent:
  • Knowledge items explicitly attached to the test (text and memory items only; links are excluded as too large)
  • Agent Visuals
  • Agent Rules matching that test (subject to targeting filters)
  • The Project Summary
Semantic search does not run at execution time. The agent works from the test’s goal, steps, and the targeted context above. Configs provide credentials, environment URLs, and test files.
This separation means you can safely update knowledge without affecting tests that are already running or scheduled.

Attaching Knowledge to a Specific Test

Library and Memory items aren’t used during execution by default. To give the agent a specific item as context while a test runs, attach the knowledge item to that test case. This is the supported way to make a Library or Memory item available during execution (text and memory items only; links are excluded as too large).

Project Knowledge vs. Chat-Local Files

Understanding the scope of your knowledge helps you organize it effectively: Project Knowledge (Settings → Knowledge):
  • Shared across all chats in the project
  • Accessible to all team members
  • Permanent and searchable by semantic search
  • Best for: Official documentation, stable specifications, product rules
Chat-Local Files (uploaded directly to a chat):
  • Per-message attachments scoped to a single conversation
  • Other users’ chats cannot access them
  • Not persisted as project knowledge - they do not automatically become reusable Library items
  • Best for: Experimental specs, PR descriptions, feature branch docs, exploratory work
This isolation prevents work-in-progress information from affecting other team members. When your experimental work becomes final, add it to the Library or ask the AI Chat to create a knowledge item from the conversation. Crawled documentation and indexed knowledge are stored in a Knowledge Graph with embeddings that enable semantic search - finding relevant information based on meaning rather than exact keywords. During chat and test creation, the assistant searches your knowledge automatically and on demand:
  1. Analyzes your request to understand intent
  2. Searches your Library, Memories, and the site/feature knowledge graph for semantically similar content
  3. Provides the most relevant information to the AI assistant
  4. Uses clear titles to improve search accuracy
The Reset Knowledge Graph control lives on the Crawling tab.

Adding Library Items

1

Navigate to Knowledge Settings

Navigate to Settings → Knowledge and open the Assistant tab.
2

Open the Library

Find the Library section and click “Add Knowledge”.
3

Choose Content Type

Choose your content type - URL or Text - based on what you want to add.
4

Fill in Details

Fill in the title and content for your knowledge item. Use clear, descriptive titles that reflect the content - this helps both the AI search and allows you to reference specific knowledge by name in chat (e.g., “use the API documentation knowledge”).
5

Save and Process

Save to process and index the content for use by AI assistants.

Organization Tips

  • Use Clear Titles: Make Library items easy to find with specific, descriptive names
  • Keep Content Current: Regularly review and update documentation links
  • Remove Outdated Info: Delete or update knowledge that no longer reflects your product
  • Test Your Knowledge: Ask the chat assistant questions to verify it finds the right information
  • Avoid Contradictions: When updating specs, remove old versions to prevent confusion

Troubleshooting

Knowledge not being found in chat:
  • Ensure content has been fully processed (check status indicators)
  • Try different search terms or questions
  • Verify the content is relevant to your question
  • Use clear, descriptive titles for better semantic search results
Rules not being applied:
  • Check that you’ve saved the rule on the correct tab (Assistant Rules, PR Review Rules, or Agent Rules)
  • For Agent Rules, verify the rule’s targeting filters (application, labels, scenario, URL path) actually match the test you’re running
  • Review test results to see whether the rule’s guidance is being followed
Summary not generating:
  • Ensure you’ve completed initial site analysis
  • Try running a few tests to build the knowledge graph
  • Manually refresh the summary after adding new functionality
Chat-uploaded files not visible to team:
  • Remember: files attached in chat are conversation-specific
  • To share with team, add the content to the Library instead
  • Other team members need to attach it to their own chats, or use project knowledge