Knowledge Base & Documents

Every project in Audra Flow has its own knowledge base — a collection of documents that your AI agents can reference when generating artifacts or answering questions. The more relevant context you provide, the more useful and accurate the agents become.

Think of the knowledge base as giving your AI agents the same background reading that a new team member would need. Business requirements, research reports, technical constraints, compliance policies, brand guidelines — anything that helps the agents understand your project's specific context.

Why Knowledge Bases Matter

Without uploaded documents, the AI agents rely only on the artifacts you have created within Audra Flow and their general training. That can be enough for simple tasks, but most real projects have important context that lives in existing documents — research findings, business rules, regulatory requirements, technical documentation from other systems.

When you upload these documents, the agents can:

• Reference specific facts, requirements, and constraints from your domain when generating content.
• Answer questions about your project using your actual documents rather than general knowledge.
• Identify gaps or inconsistencies between what your documents say and what your artifacts cover.
• Produce specifications, user stories, and technical plans that reflect your real business context.

The difference is significant. An agent generating a product specification with access to your compliance documents, brand guidelines, and stakeholder interviews will produce something far more relevant than one working without that context.

Uploading Documents

You can upload documents to your project from the Knowledge Base section in the project sidebar.

How to Upload

1. Open your project and navigate to the Knowledge Base section in the sidebar.
2. Click the Upload Document button.
3. Select one or more files from your computer. You can also drag and drop files into the upload area.
4. Wait for the upload and processing to complete. A progress indicator will show you the status of each file.
5. Once processing finishes, the document appears in your knowledge base and is immediately available to all AI agents in the project.

Supported Formats

There is no strict limit on the number of documents you can upload, but keeping your knowledge base focused on relevant materials will give better results than uploading everything you have.

How Documents Are Processed

When you upload a document, Audra Flow processes it in the background so the AI agents can work with it efficiently. Here is what happens:

1. Text extraction: The content of your document is extracted. For PDFs, this includes text from all pages. For Markdown and plain text, the content is read as-is. For Office documents, formatted content is converted to searchable text.
2. Chunking: The extracted text is divided into smaller, manageable sections. This allows the agents to find specific, relevant passages rather than working with entire documents at once.
3. Indexing: Each section is analyzed and indexed so that when an agent needs information, it can quickly find the most relevant parts of your documents.

The end result: when you ask a question or generate an artifact, the AI agent searches your indexed documents, finds the passages most relevant to your request, and uses them as context for its response. This means the agent's output is grounded in your actual project materials, not just general knowledge.

Processing typically takes a few seconds for small documents and up to a minute for longer files. You will see a status indicator while processing is underway.

Asking Questions with Product Guru

The Product Guru agent is specifically designed to query your knowledge base. While all four AI agents can reference your uploaded documents when generating artifacts, the Product Guru is purpose-built for finding information and answering questions about your project materials.

How to Ask Questions

1. Open the Product Guru panel from the project sidebar or the chat panel.
2. Type your question in natural language. You do not need special syntax or keywords.
3. The Product Guru searches your knowledge base, finds relevant passages, and provides an answer with references to the source documents.

Example Workflows

Here are some common ways teams use the Product Guru with their knowledge base:

• Quick fact-finding: “What is the maximum response time requirement mentioned in the SLA document?”
• Document summaries: “Summarize the key findings from the user research report.”
• Cross-document analysis: “Do our technical requirements document and our compliance policy mention the same data retention period?”
• Gap identification: “What topics are covered in our business requirements but not addressed in any of the technical documents?”
• Onboarding new team members: “Give me an overview of the project based on the uploaded documents, including the main goals and key constraints.”

The Product Guru always tells you which documents it used to form its answer, so you can go back to the source material and verify the details.

Managing Documents

Your knowledge base is not a one-time setup. As your project evolves, you will want to add new documents, replace outdated ones, and keep things organized.

Viewing Your Documents

The Knowledge Base section shows all uploaded documents with their file name, upload date, file size, and processing status. You can see at a glance which documents are fully indexed and available to the AI agents.

Organizing Your Knowledge Base

While there are no folders or tags within the knowledge base, you can keep things manageable by:

• Using clear, descriptive file names before uploading (for example, “Q3-2025-User-Research-Report.pdf” is more helpful than “report-final-v2.pdf”).
• Removing documents that are no longer relevant to avoid cluttering the agents' search results.
• Uploading updated versions of documents when the content changes, rather than keeping multiple outdated copies.

Removing Documents

To remove a document from your knowledge base, find it in the Knowledge Base section and click the delete option. The document and its indexed data will be removed. The AI agents will no longer reference it in future responses.

Removing a document does not affect any artifacts that were previously generated using that document's content. Those artifacts remain as they are — only future AI interactions will no longer reference the removed document.

Per-Project Isolation

Every project's knowledge base is completely isolated from every other project. This is a fundamental part of how Audra Flow handles your data.

• Documents uploaded to Project A are not visible, searchable, or accessible from Project B — even if both projects belong to the same organization or team.
• AI agents working within a project can only reference that project's knowledge base. They cannot pull information from other projects.
• If you need the same document in multiple projects, upload it separately to each one. This ensures that each project's knowledge base is a deliberate, curated collection.
• Deleting a project removes all of its knowledge base documents and indexed data permanently.

This isolation ensures that sensitive information in one project — such as financial data, legal documents, or proprietary research — cannot leak into another project, even unintentionally.

Best Practices

Getting the most from your knowledge base comes down to giving the AI agents the right context at the right time. Here are some practical recommendations.

Upload Early

Add your key documents before you start creating artifacts. If the agents have your research reports, business requirements, and constraints available from the beginning, every artifact they help generate will benefit from that context. Uploading documents after you have already created most of your artifacts means those earlier artifacts were generated without the full picture.

Keep Documents Current

If a business requirement changes or a research finding is updated, upload the new version and remove the old one. Outdated documents can lead the agents to reference information that is no longer accurate. The agents do not know which version of a document is the “latest” — they treat everything in the knowledge base as current and authoritative.

Include Diverse Sources

The more perspectives your knowledge base covers, the richer the agents' output will be. Consider uploading:

• User research: Interview transcripts, survey results, usability test reports.
• Business context: Requirements documents, project charters, stakeholder presentations, strategy decks.
• Technical context: Existing architecture documents, API documentation, integration guides, system constraints.
• Compliance and policy: Regulatory requirements, security policies, accessibility guidelines, data governance documents.
• Competitive intelligence: Market analysis, competitor feature comparisons, industry benchmarks.

Be Selective

Quality matters more than quantity. Uploading hundreds of marginally relevant documents can actually reduce the quality of the agents' output, because they have more noise to search through. Focus on documents that contain information the agents will actually need when generating artifacts and answering questions for this specific project.

Use Descriptive File Names

When the Product Guru references a document in its answer, it shows the file name. Names like “User-Research-Q3-2025.pdf” or “Payment-Integration-Requirements.md” make it easy for your team to know exactly which source is being cited. Names like “doc1.pdf” or “final-FINAL.docx” do not.

Next Steps

• Learn how the AI agents use your knowledge base in the Working with AI Agents guide.
• Start the Discovery phase to see how the UX Researcher draws on your uploaded research documents.
• Read about Roles & Permissions to understand who on your team can upload and manage documents.