Marketplace›AI and automation›AI development and chatbots›RAG chatbot that answers from your documents and shows sources
RAG chatbot that answers from your documents and shows sources

RAG chatbot that answers from your documents and shows sources

AI development and chatbots · delivered in 5-21 days

Customers and staff often ask questions your own files already answer, yet a generic chatbot will guess rather than admit it doesn't know. We build you a chatbot that searches your documents first and answers strictly from what it finds, pointing to the exact file and page or section, and flagging clearly when nothing relevant exists.

  • Starter: up to 50 files, source-cited answers, tested against up to 20 of your questions, source code included, 1 revision
  • Standard: up to 300 files, source-cited answers, tested against your questions, source code included, 2 revisions
  • Advanced: up to 2,000 files, deployable inside your Postgres or Supabase application or on your own server with a local model, source-cited answers, tested against your questions, source code included, 3 revisions

Your answers are usually already written down somewhere: policies, manuals, price sheets, contracts, help articles. We build a chatbot that reads those files first and replies only from what it locates there, naming the file and page or section behind each reply. If the material doesn't cover a question, the bot says so rather than inventing something.

Each package closes with a check against up to 20 of your own questions. You receive a table showing every answer, the source behind it and whether it was correct; from there we apply one round of corrections and rerun the test. You keep the source code outright, and the assistant operates through your own account with your chosen AI provider, so usage appears directly on their invoice and we charge no ongoing fee.

Search is handled by a compact model running alongside the bot, with the index stored in PostgreSQL using pgvector, meaning your documents are not sent to the AI provider, and any Postgres or Supabase setup needs no separate vector database. Recent example work includes building the vector search layer for a live Supabase backend used by an election trading desk, featuring filtered similarity search, access limited to specific teams, and hourly summaries generated by the language model and reviewed before being saved.

For 14 days following delivery, we correct anything that does not match what was agreed, at no extra cost.

There are no calls involved. We agree the scope in writing before starting, and you receive a written progress note at the close of each working day.

What happens after you buy

Your order opens its own thread here the moment it is paid, and everything about that order - questions, changes and the final report - happens in it.

Questions people ask

Will it make answers up?

The bot replies only using passages it actually retrieves, naming the file and page or section each time. When nothing relevant is found, it tells you directly. The test run against your own questions lets you judge its accuracy before accepting the work.

Which files can you use?

We work with text-based PDFs, Word documents, plain text, Markdown and public web pages; scanned pages require the optional text recognition add-on. A single file covers up to 50 pages, or 25,000 words where there are no pages, and longer files are counted in blocks of 50 pages.

Who pays for the AI and hosting?

You cover these costs yourself: the bot runs under your own API key on a host registered to you, so no monthly charge applies to us. Starter includes instructions so you can run it; with Standard and Advanced we configure it on your hosting. Running costs depend on how much traffic it gets.

Can it go inside my existing app?

Yes, this is available with Advanced. Using PostgreSQL or Supabase, the search index sits inside your own database, with your access rules controlling which documents each user can see. If you use a different stack, get in touch before placing an order.

Is my data private?

Search is powered by a small model that runs together with the bot, so your files never reach the AI provider; only the matching passages are sent per question. With Advanced, answers can instead come from a local model on your own server, provided it has sufficient memory.

People also buy

AI development and chatbots

AI agent and custom LLM integration for business workflows

We build and integrate AI agents and custom LLMs into your workflows, automating support, lead handling, data extraction and content tasks so your team can focus on higher-value work.

$375.99 - $1,874.99PACKAGES FROM 3-8 days
AI development and chatbots

AI automation and business systems that save 10+ hours a week

An AI automation and business systems specialist helps small business owners and operations teams reclaim 10 to 20 hours a week using chatbots, workflow automation and dashboards, backed by a track record across more than 50 delivered projects, offered through PBN.LTD as fixed-price builds.

$93.99 - $3,125.99PACKAGES FROM 2-10 days
AI development and chatbots

AI chatbot training and response optimization

PBN.LTD reviews and refines your existing AI chatbot, improving prompts, FAQ handling and conversation flow so it responds more naturally and supports customers better.

$12.99 - $118.99PACKAGES FROM 1-5 days
AI development and chatbots

AI voice receptionist for your business

A custom-built AI voice receptionist that answers calls around the clock, books appointments and logs customer details automatically for clinics, salons and service businesses.

$62.99 - $437.99PACKAGES FROM 3-7 days