RAG Integration

The RAG integration lets your AI assistant search and answer questions from your own documents. Upload markdown files once, and every query against this tool searches across them and returns the most relevant passages, with the source document named alongside each result.

Capabilities

• Document Search: Turns a natural-language question into a search across your uploaded documents and returns the most relevant passages.
• Source Attribution: Every result comes back together with the name of the document it was found in, so answers can be traced back to their source.
• Self-Contained Knowledge Base: Each tool has its own private set of documents — nothing is shared across tools or projects.

Integration

The RAG integration has no external service to connect — there's no host, credentials, or connection string to configure. Instead, you build its knowledge base directly by uploading documents.

Uploading Documents

Configure the tool's knowledge base by uploading the files you want the AI to be able to search.

Requirement	Details
File Format	Markdown only (.md or .markdown).
Max File Size	20 MB per file.
Number of Files	Multiple files can be uploaded at once.

Drag and drop your files into the Documents step when creating or editing the tool, or use Browse Files to select them. Each file then appears in the documents table with a processing status:

• Pending: The file has been received and is queued for processing.
• Processing: The file's content is being indexed for search.
• Ready: The document is indexed and available to be searched.
• Failed: Indexing failed; check the error message shown next to the document, then fix the issue and re-upload it.

Re-uploading a Document

Uploading a file with the same name as an existing document replaces its content and re-indexes it — you don't need to delete the old one first.

Processing Takes a Moment

Newly uploaded documents move from Pending to Ready shortly after upload. Use the refresh control in the documents table to check on progress, and wait for a document to reach Ready before relying on it in a search.

You can add, update, or remove documents at any time from the tool's edit view — changes take effect immediately, with no redeploy needed.

Configuration

Once your documents are uploaded and marked Ready, the tool itself needs no further configuration. It exposes a single input to the AI assistant:

Parameter	Required	Description
Query	Yes	The natural-language question or search phrase the AI is looking to answer.

When called, the tool searches across all Ready documents for this tool and returns the most relevant passages it finds, each labeled with the name of its source document. If nothing relevant is found, the tool reports that no matching documents were found rather than guessing at an answer.

Keep Documents Focused and Well-Structured

Search quality depends on how your documents are written. Break large topics into clearly headed sections rather than one long unstructured file — this makes it far more likely the AI retrieves exactly the passage it needs.

Security & Patterns

This section outlines the guardrails and usage patterns that keep a RAG tool's knowledge base safe and effective in production.

Document Isolation

Documents are scoped to the specific tool you upload them to. A document uploaded to one tool is never visible to, or searchable from, another tool or project.

Treat Uploaded Documents as Shared Context

Any document you upload becomes searchable by anyone who can invoke this tool. Don't upload documents containing information you wouldn't want surfaced in an AI response — for example credentials, secrets, or regulated personal data.

Content Guidelines

• Only markdown files are accepted; convert other formats (PDF, Word, HTML, etc.) to markdown before uploading.
• Keep documents current. Outdated content is returned with the same confidence as up-to-date content, since the tool has no way to know a document is stale.
• Remove documents you no longer want referenced. Deleting a document also removes it from future search results.

Handling Failed Documents

If a document's status shows Failed, review the error message shown in the documents table, fix the underlying issue (for example, an unsupported file type), and re-upload the file. A failed document is never included in search results.

Performance Optimization

Pattern	Description	Benefit
Focused Documents	Split broad topics into smaller, single-purpose documents.	Improves retrieval precision.
Clear Headings	Structure content with descriptive headings and short sections.	Helps the AI find the passage that answers the question.
Regular Cleanup	Remove outdated or duplicate documents.	Reduces noise in search results.

Protocol Limitations

• Only markdown documents are supported; other file types are rejected at upload.
• Very large documents may take longer to move from Pending to Ready.
• The tool returns text passages only — it does not render images, tables, or other embedded media from the source document.