Sold by Zinn Digital
Get a custom chatbot that answers strictly from your documents with exact source citations, and admits openly when information is missing.
Your crucial answers are already saved across your manuals, policies, price lists, help articles and contracts. We create a chatbot that digests these documents and responds exclusively using the data it finds within them, providing the exact file and page or section reference for every response. When an answer is missing from your files, the assistant states this clearly instead of guessing.
Every project concludes with a verification test using up to 20 of your own questions. You get a detailed spreadsheet listing each reply, its source and its accuracy rating, followed by one revision round and a complete retest. You retain full ownership of the source code, and since the bot runs on your personal account with your selected AI provider, all usage bills go directly to you with zero recurring fees from us.
Document search runs via a compact model paired with the bot, while the index is housed in PostgreSQL using pgvector. This keeps your files away from the AI provider and eliminates the need for an external vector database on any Postgres or Supabase installation. Recent work includes developing the vector search system for a live Supabase backend on an election trading desk, which included filtered similarity search, team-based access restrictions and language model generated hourly summaries approved prior to saving.
We provide 14 days of post-delivery support to fix any discrepancies against the agreed specifications at zero additional charge.
There are no video or phone calls required. All project details are agreed in writing beforehand, and you receive a written update at the end of every working day.
$112.99
Or pay €100.85 · £84.45 at today's rate — the rate is locked when you add it to your basket.
A web chat on up to 50 of your files, citing sources, tested on up to 20 of your questions. - Up to 50 files - Answers with sources - Test on your questions - Source code - 1 revision
Delivered in 5 days
Revisions included: 0
$312.99
Or pay €279.36 · £233.92 at today's rate — the rate is locked when you add it to your basket.
- Up to 300 files - Answers with sources - Test on your questions - Source code - 2 revisions
Delivered in 10 days
Revisions included: 0
$687.99
Or pay €614.07 · £514.19 at today's rate — the rate is locked when you add it to your basket.
Up to 2,000 files, in your Postgres/Supabase app or on your server with a local model. - Up to 2,000 files - Answers with sources - Test on your questions - Source code - 3 revisions
Delivered in 21 days
Revisions included: 0
No account needed first — yours is created at checkout. We hold the money until you accept the work.
The bot responds solely with retrieved passages, citing the specific file and page or section every time. If no matching information exists, it states so plainly. The verification test using your own questions allows you to evaluate its precision prior to sign-off.
We accept Word documents, text-based PDFs, Markdown, plain text and public web pages; image-based scans need the optional text recognition add-on. One file accommodates up to 50 pages or 25,000 words for unpaginated text, with longer documents calculated in 50-page blocks.
You pay these expenses directly: the assistant operates via your API key on your own hosting account, meaning we charge no monthly fees. The Starter package includes setup instructions, while Standard and Advanced include our configuration on your server. Ongoing costs scale with your usage volume.
Yes, this feature comes with Advanced. By utilising PostgreSQL or Supabase, the search index resides within your database, using your permission controls to dictate document visibility per user. If your tech stack differs, please contact us before ordering.
Retrieval is handled by a lightweight local model running alongside the bot, meaning your files are never transmitted to the AI provider; only the specific matching passages are sent per query. With Advanced, responses can alternatively be generated by a local model hosted on your own server, provided it possesses adequate memory.