Knowledge Bases

Knowledge Bases

Knowledge bases give your AI agents access to your content. Instead of relying on generic training data, agents can reference your documentation, product specs, support articles, and style guidelines to produce accurate, on-brand responses.

How It Works

When an AI agent receives a query, Outrun searches your knowledge bases for the most relevant content and injects it into the agent's context. The agent then uses that context to generate its response, grounding its answers in your actual content rather than guessing.

This retrieval-augmented approach means your agents stay accurate and up to date as your content changes — no retraining required.

Types of Knowledge Bases

Outrun supports two types of knowledge bases, each serving a different purpose:

Knowledge Groups

Knowledge groups are collections of documents that agents search at query time. When a visitor asks a question, Outrun finds the most relevant passages from your documents and provides them to the agent as context.

Use knowledge groups for:

• Product documentation and help articles
• Internal process guides and runbooks
• Technical specifications and API references
• FAQs and support content

Style Guides

Style guides define the tone, voice, and writing style your agents should follow. Rather than searching for relevant passages, the entire style guide is injected into the agent's instructions.

Style guides can be created in two ways:

• Auto mode — Upload example content (emails, support replies, marketing copy) and Outrun distils a style guide from the patterns it finds
• Manual mode — Write your style rules directly, specifying tone, vocabulary, formatting preferences, and any other guidelines

Use style guides for:

• Brand voice consistency across agents
• Tone guidelines (formal, casual, technical)
• Vocabulary and terminology standards
• Response formatting rules

Creating a Knowledge Base

1. Navigate to AI > Knowledge in your workspace
2. Click Create Knowledge Base
3. Choose the type: Knowledge Group or Style Guide
4. Give it a name and optional description
5. Add your content (see below)

Adding Content

Uploading Documents

You can upload files directly to a knowledge group. Supported formats include:

• Markdown (.md)
• Plain text (.txt)
• HTML (.html)
• JSON (.json)
• Code files (.js, .ts, .py, .rb, and more)

Uploaded documents are automatically split into chunks, embedded, and indexed for semantic search.

Syncing from Sources

If you have a connected source (like a GitHub repository), you can sync content directly from it. Outrun pulls in files, processes them, and keeps the knowledge base updated when the source changes.

Configuration Options

System Prompts

Each knowledge base can have a system prompt — a set of confidential instructions that tell the agent how to use the content. System prompts are injected into the agent's context but are never shown to end users.

Use system prompts to:

• Tell the agent which topics this knowledge base covers
• Set boundaries on what the agent should and should not answer
• Provide context about the audience (technical users, customers, internal team)
• Define escalation rules (when to hand off to a human)

Source URL Prefix

Set a source URL prefix to link agent responses back to the original documents. When an agent cites a passage, it can include a link to where that content lives — for example, a page in your documentation site or a file in your GitHub repository.

This is especially useful for customer-facing agents where users may want to read the full source material.

Folder Filtering

For knowledge bases synced from large repositories, use folder filters to scope the content to specific directories. This keeps the knowledge base focused and prevents irrelevant files from diluting search results.

For example, if you sync a GitHub repository, you might filter to only include the docs/ directory rather than indexing the entire codebase.

Connecting Knowledge Bases to Agents

Knowledge bases are attached to AI agent nodes in your workflows:

1. Open the workflow editor and select an AI agent node
2. In the node settings, find the Knowledge Base section
3. Select one or more knowledge groups and optionally a style guide
4. Save the workflow

When the workflow runs, the agent automatically retrieves relevant context from the selected knowledge bases before generating its response.

You can attach multiple knowledge groups to a single agent. Outrun distributes the search across all of them and merges the most relevant results.

Best Practices

Keep documents focused

Smaller, topic-specific documents produce better search results than large catch-all files. If a single document covers many unrelated topics, consider splitting it into separate files.

Update content regularly

Knowledge bases reflect whatever content you have uploaded or synced. If your product, policies, or processes change, update the corresponding documents so your agents stay current.

Use folder filters for large repos

When syncing from a source with thousands of files, folder filters prevent noise from irrelevant content. Point the knowledge base at just the directories that matter — like docs/, help/, or guides/.

Write clear system prompts

A good system prompt tells the agent exactly how to use the knowledge base. Be specific about scope, audience, and tone. For example: "Use this knowledge base to answer questions about our billing plans. If the user asks about a technical integration, tell them you can help with billing questions and suggest they check our integration docs."

Combine knowledge groups with a style guide

Attach a style guide alongside your knowledge groups to ensure the agent not only has the right information but delivers it in the right voice. This is especially important for customer-facing agents where brand consistency matters.

Review agent responses

Use the human-in-the-loop feature to review agent responses and provide corrections. Outrun learns from your feedback over time, improving retrieval accuracy and response quality for similar queries in the future.