About Velqor - Private AI Document Search (Source Code)
Self-hosted AI search with cited answers over company files. Full source code.
Velqor is a complete, tested codebase for a private "company memory": a business uploads its documents (or syncs a Google Drive folder), then searches them in plain language and asks questions. Answers are written by a local AI model, and every statement links to the passage it came from. It runs entirely on one machine with open-source parts. No OpenAI or other paid API is required, and no document leaves the server. Optional adapters for Anthropic and OpenAI are included and switched off by default. This is a source code sale, not a running business. There is no hosted service, no customers and no revenue. You get the full codebase, documentation and brand assets, and you can launch it, white-label it, or use it inside your own company. Sold once, to one buyer: all rights transfer to you exclusively, and the code is not sold to anyone else or kept in use by the seller. What it does: • Upload files or whole folders: PDF (with OCR for scans), Word, Excel, PowerPoint, CSV, text. Type, language, dates, amounts, companies and reference numbers are detected automatically. • Hybrid search (English and French full text, vectors, file names, reference numbers, companies), with plain-language reasons for every result. • AI answers that stream in, cite their sources, and are checked: statements that cannot be matched to a source are removed; numbers must appear in the cited document. • Summaries (short, detailed, action items) and version comparison that finds changed terms (for example payment terms 30 → 45 days) without AI. • Organizations, roles, invitations, groups, per-document sharing, audit log, admin dashboard, data export, organization and account deletion. • Google Drive connector: read-only, incremental sync, sharing mirrored from Drive. Built for trust: • Permissions are applied inside every database query, plus PostgreSQL row-level security as a second wall. • Append-only audit log, encrypted connector tokens, strict Content Security Policy, rate limits. • 360+ automated tests, a 28-step end-to-end test on the production build, and evaluation sets for search and answers (0 leaks of restricted documents). • Search over 20,000 passages in about 0.2 seconds. Documentation included: architecture, data model, 36 recorded design decisions, security model with per-phase checklists, evaluation results, operations guide (Docker, backups, updates), and a buyer handover guide with an honest list of known gaps.
Project details
Website: https://claude.ai/artifact/P3Egr9aXSbeAZsDDoagnQU
Next step: <p>• Launch it as a hosted product for small and mid-sized companies that cannot send their documents to US AI providers. Add hosting, email sending and billing; organizations, roles and usage metering are already built in.</p><p>• Sell it as an on-premise install to firms with strict confidentiality (accountants, law firms, engineering offices), with setup and support fees.</p><p>• White-label it for IT service providers who want a private AI search offer for their clients.</p><p>• Add more sources: the connector design is ready for OneDrive/SharePoint, Outlook and Gmail.</p><p>• Improve answers with a larger local model, or switch on the included Anthropic or OpenAI adapter.</p><p>• Use it internally as a private AI search for your own company.</p>
For sale
Asking price: $7,500 USD
What's included
Source code, All related design work
• Full source code, delivered after payment as a clean Git repository with one commit (no history, no secrets). • Documentation: architecture, data model, 36 design decisions, security model with checklists, quality results, operations guide (Docker, backups, updates), and a buyer handover guide listing known gaps. • Test suite (360+ tests, end-to-end test) and evaluation sets for search and answers. • Container files (Dockerfile, Docker Compose) and a demo script that creates a sample company in one command. • Brand assets: name, logo, colours and font setup. • Exclusive, full transfer of all rights to one buyer by written agreement. • 14 days of support after the sale, for setup and questions about the code.
Reason for selling
I built Velqor as a complete product, but I don't have the time to take it to market and support customers myself. It deserves an owner who will launch it.