Vector Database Search Agent - Free Preview

## Key Capabilities

• Conversational Agent (POST /agent): One natural-language interface for everything - the agent decides whether to search your documents, add a new one, or list your collections based on what you ask. Remembers conversation context automatically across requests.
• Document Provisioning: Add documents directly via API (paste text) or upload existing files - .txt, .md, .pdf, .docx - which are extracted and automatically chunked server-side, no pre-processing required.
• Semantic Search: Query in plain language and get back the most relevant excerpts, with source file name, type, chunk position, and relevance score.
• Collection Management: List or delete your own collections at any time via direct API calls.
• * Self-Service Data Control: Export a full copy of your data (account info, documents, conversation history) or delete your account and everything in it at any time via the API - no waiting on support. Unsubscribed accounts are automatically and permanently deleted within 30 days.

## Supported Document Formats

PDF, DOCX, Markdown, and plain text files (or paste raw text directly via the API).

## How It Works

Documents are embedded using ChromaDB's built-in sentence-embedding model (all-MiniLM-L6-v2) and stored in your own private vector index. Uploaded files are automatically split into overlapping chunks at natural word boundaries so long documents don't need manual pre-processing. Queries are embedded the same way and matched by semantic similarity, returning ranked excerpts with source metadata.

The conversational agent layer is built on Google's Agent Development Kit (ADK). You supply your own Google Gemini API key when calling the agent endpoint - it's used only for that request and never stored, so any LLM usage cost is billed to your own Google account, not ours. Direct search/document endpoints (no agent) require no LLM key at all.

## Use Cases

• Internal Knowledge Search: Index company documents and ask natural-language questions to retrieve relevant passages from reports, proposals, and meeting notes.
• Retrieval-Augmented Generation (RAG): Use the search or agent endpoints as a retrieval component in your own RAG pipeline.
• Research and Technical Documentation Search: Search across PDFs, DOCX files, and Markdown docs by meaning, not exact wording.
• Conversational Document Q&A: Attach a file to the agent and ask questions about it in the same request - one call does upload, index, and answer.

## Who Is This For?

• AI and data teams building RAG workflows
• Developers integrating document search or a conversational agent into their own app
• Research and business teams needing faster discovery across document repositories

## Technical Details

• Vector Database: ChromaDB, local persistent storage, one isolated index per customer
• Embedding Model: all-MiniLM-L6-v2
• Agent Framework: Google ADK, buyer-supplied Gemini API key per request
• Chunking: Fixed-size, word-boundary-aware, 1000 characters with 100-character overlap (automatic on file upload)
• Default Top-K Results: 5 (maximum 20)

## Metadata Captured

File name, file type, and chunk index are returned with every search result for traceability back to the source document.

## Security & Compliance

Your data is encrypted at rest and in transit (HTTPS/TLS). Dependencies are scanned daily for known vulnerabilities, with continuous monitoring via GitHub Dependabot. Every customer's data is fully isolated - never shared or visible across accounts. Full details, including our data retention/deletion policy, are in our Privacy & Data Protection Statement and End User License Agreement.

## Getting Started

• Subscribe to the free product through AWS Marketplace.
• You'll be redirected automatically to receive your API key.
• Add your documents: paste text via the API, or upload files directly.
• Search directly via API, or talk to the agent in natural language (bring your own Gemini API key for the agent endpoint).
• Review retrieved excerpts and source metadata, or use the agent's natural-language answers directly in your workflow.

## Product Differentiation

Vector Search API combines document ingestion, automatic chunking, isolated per-customer vector indexing, and both direct-API and conversational-agent access in one product. Unlike simple keyword search, it retrieves relevant content even when the query doesn't match the exact wording in your documents, and the agent interface means integrators don't need to learn multiple endpoints to get value.