AI chatbot for Airbnb hosts

No. Agentic safeguards are in place to ensure no hallucinations occur. If AI cannot find the answer in the provided context from your collection, rather than invent an answer it responds with “The answer is not contained in the provided context.”

AskBnB uses the highest security standards: SOC 2 Type II compliant, HIPAA compliant and GDPR compliant.

• End-to-end encryption. Any data in motion is encrypted end to end.
• Google Cloud Storage. For data at rest, volumes are stored on Google Cloud Storage using Google's security protocols.
• Data privacy. We do not sell your data or use it for any purpose other than providing the AskBnB service. We do not train models on your data.
• Access control. There is an optional access code requirement if you prefer guests to enter a code before chatting with a collection you have shared publicly to a dedicated URL.

No. AskBnB is browser based and mobile friendly, so when guests scan your QR code they don't have to download an app or register. They can start asking questions immediately.

AskBnB has special handling for the broadest range of file format support of any RAG platform in the marketplace. That includes, but is not limited to, Google files (Docs, Sheets, Slides), MS Office files (.docx, .xlsx, .pptx), images, video files, audio files, CSV, multi-tab XLSX files, PDF, markdown and more.

No.

Yes. In Settings › Preferences you can set any LLM model you would like. The default is Auto, where AskBnB makes the selection for you, but you can change it to your preferred model — including the latest from ChatGPT, Gemini, Grok and more. New models are added as they are released.

This integration keeps files in your Google Drive or MS OneDrive auto-synced with collections on the AskBnB platform, so any change to a document is synced every 24 hours.

Auto-sync: while choosing a file from Google Drive or MS OneDrive, tick the Auto-sync checkbox before clicking Vectorize. AskBnB monitors the synced file for changes and updates your collection every 24 hours. Your AI always has the latest information and you never have to upload the same document twice.

Overview. RAG is a technique that enhances large language models by allowing them to retrieve and use information that was not included in their training data. While an LLM's knowledge is fixed at its training cutoff, RAG enables access to personal, proprietary or up-to-date data at query time.

How RAG works. RAG processes your documents by converting them into tokens and vector embeddings — numerical representations that capture semantic meaning. When a user asks a question, the system retrieves the most relevant document chunks and supplies them to the LLM as context before generating a response.

When a RAG platform is useful. When you need the AI to know information it was not trained on; when you work with large or numerous documents that exceed the model's context window; when you require long-term memory over large knowledge bases; when you use file formats LLMs don't natively support; or when you want to expose your domain knowledge through a queryable AI interface.

Why RAG is powerful. It performs semantic similarity searches across vectorized documents, identifies the most relevant content in high-dimensional vector space, supplies only the most pertinent information to the LLM, and reduces hallucinations by grounding responses in your data.

The result. Highly targeted, accurate responses even across massive document collections. Well-defined, specific questions produce the best results, as they map more effectively to the vectors containing the answer.

Same models, different knowledge sources. They are the same models but draw on different data. The regular ChatGPT you use directly answers from public internet data it was trained on. The AskBnB platform augments that model with the data you vectorized into your collection — and in this case the model must ground its answer in the source data you provided.

What makes this unique. This is what turns a model like ChatGPT into your own personal LLM — an expert on anything you want it to know about your short-term rental and neighbourhood.

When you ask about data you vectorized into your AI library, the model does not attempt to fetch the answer from its generalized pretraining data. It fetches the answer from your source material. If the answer doesn't exist in your library, it says so instead of hallucinating.

The real power is in the vector embeddings, vector database and retrieval techniques that fetch only the relevant parts of any document as context — even if that information is buried across hundreds or thousands of documents.

Platform capabilities. Curate your AI library from any source, in almost any format, across hundreds to thousands of documents, videos, audio files, web URLs, social posts and images that you or anyone else can then query. When AI answers, links to the source material are provided so you can see the plain text parsed from the original file or URL. That feature is turned off for any collection shared to a dedicated URL, so guests cannot access your source material.

Pre-processing and post-processing are the secret sauce of a high-performing RAG system.

Pre-processing. Before we vectorize your data, we clean and structure it to maximise retrieval accuracy:

• Context continuity. Breaking documents into chunks can leave large context holes without sophisticated techniques and agentic workflows to maintain continuity across vectors. Different sources need different handling — chunking a video transcript is not like chunking a spreadsheet.
• Structure retention. We preserve the original document's structure (headings, lists, tables) in a format the LLM understands (markdown). This is crucial for understanding the relationships between parts of a document.
• Metadata addition. We add metadata (source URL, document type, date) to each chunk, which allows powerful filtering during retrieval.

Post-processing. After retrieving the most relevant chunks we refine them further:

• Re-ranking. Additional algorithms re-order retrieved chunks by relevance.
• Vector overlap removal. Redundant information between chunks is eliminated.
• Chunk stitching. Related chunks are combined into cohesive context.
• Contextual irrelevance QA. Agentic workflows discard retrieved chunks that look relevant by vector similarity but aren't actually relevant to the query.
• Metadata injection. Relevant metadata is included in the context passed to the LLM.

Why it matters. Without proper pre- and post-processing, RAG systems often return irrelevant or incomplete results. Our expertise here is what makes AskBnB's retrieval accurate and reliable.

AskBnB uses specialised handling for tabular data across multiple tabs.

• Structure preservation. The table structure is retained as structured markdown before the vector embedding step, so the LLM understands the relationships between rows and columns.
• Multi-tab support. Multi-tab XLSX files are handled with the structure of each tab preserved — crucial for complex spreadsheets organised across multiple sheets.
• Querying tables. You can ask questions that require the LLM to analyse and synthesise information from tables, just like any other data type.