About this software
About Assistlyai
Assistly AI is a multi-tenant Next.js 15 / React 19 / TypeScript app backed by SQL Server through Prisma. Businesses embed a one-line, plain-JavaScript widget (Shadow DOM, domain-locked per website) and visitors get answers from Claude, grounded in that website's own knowledge base of FAQs, crawled pages, PDF and Word files. Agents watch chats in a live Inbox, take over from the AI, hand back, assign or close, and leads and analytics are collected per website. A Super Admin role manages companies and platform-wide AI, email and demo settings. The package includes a landing page, install guides for VPS, Docker and IIS, SQL scripts, and a test suite.
Why Assistlyai
What makes it better
than the alternatives
- 01
Knowledge-grounded answers
Each website has its own knowledge (FAQs, crawled pages, PDF/Word uploads) and the AI answers from it, with an AI key and model configurable in Settings.
- 02
Live takeover in one widget
Agents take over from the AI in the Inbox, with long-poll delivery of messages to the visitor in the same conversation, plus hand-back and assignment.
- 03
Multi-tenant with tested isolation
Company, website and role scoping (Super Admin, Admin, Agent), with tests that fail on unguarded routes/actions and on cross-tenant access.
- 04
Cost and abuse controls
Configurable AI reply limits per website, company, chat, visitor IP per day and platform per day, plus domain lock, rate limiting and CSRF/same-origin checks.
- 05
Ready to self-host
Dockerfile, production and dev compose files, nginx and IIS configs, health endpoint, backup/restore tools, generated SQL scripts for SSMS, and CI running lint, typecheck and tests.
Chat Widget
- One script tag with a data-key
- position and mode options
- Shadow DOM styling isolated from the host page
- Domain lock so it only appears on allowed domains
- Typing indicator
- history after reload
- end chat
- mobile full-screen layout
- Arabic right-to-left support covered in browser tests
- Browser end-to-end test suite (Playwright
- 32 checks) with a mock server
AI Answering
- Claude-based replies grounded in the website's knowledge
- with sources
- Says so and offers a person when unsure
- Configurable model (Haiku
- Sonnet
- Opus) and thinking effort
- Monthly
- per-chat
- per-IP and platform daily reply limits
- Suggested replies endpoint for agents
Knowledge Base
- FAQs
- website crawling and PDF/Word file uploads (10 MB max)
- Safe-fetch protection for crawling
- tested
- Text extraction and a searchable index with caching
- Per-website knowledge management and training pages
Inbox and Live Takeover
- Live conversation list with filters and visitor panel
- Reply box
- take over
- hand back to AI
- assign to a colleague
- close
- Alerts when a visitor asks for a person
- Optional email notification to team members
Leads
- Contact capture of name
- email and phone when a person is needed
- Per-website lead list with deletion
- CSV export
Analytics
- Chats
- AI-resolved rate
- hand-offs
- first-response time
- leads and AI usage
- Per-day charts and per-website views
- CSV export
Websites and Install
- Multiple websites per company with public keys and allowed domains
- Install page with snippets
- install checker
- QR code and send-to-developer form
- Hosted chat page at /
c/[key] - Customisable widget settings
Administration and Security
- Roles: Super Admin
- Admin
- Agent
- with team and per-website access
- Company management and platform settings for AI
- email and demo account
- Hashed session tokens
- password reset by email
- audit log
- JSON/SQL Server backup
- restore and website export tools
- Demo account seeding with sample data
| Language | Version | Framework | Used for | Share of code |
|---|---|---|---|---|
| TypeScript | — | — | Front-end | 42.6% |
| TypeScript (React) | — | — | Front-end | 22.3% |
| CSS | — | — | Styling | 16.6% |
| HTML | — | — | Front-end | 6.3% |
| JavaScript | — | — | Front-end | 6.2% |
| SQL | — | — | Database | 4.8% |
| PowerShell | — | — | Scripts | 0.9% |
| Python | — | — | Web, API | 0.4% |
- Primary language
- TypeScript 5.9 on Node.js 22+
- Framework
- Next.js 15 (App Router) + React 19, Prisma 6
- Database
- SQL Server 2019+ via Prisma (SQL Express and Azure SQL work)
- Authentication
- Email and password login; random session tokens stored hashed in the database; password reset by emailed link; role checks on every route and action
- Licence
- Single-site
- Screens
- 20
- Database tables
- 16
- Supported languages
- English interface; Arabic supported in the chat widget with right-to-left layout
- Browser support
- Modern evergreen browsers (Chromium, Firefox, Safari, Edge); widget browser-tested in Chromium
| Minimum | Recommended | |
|---|---|---|
| CPU | 2 vCPU | 4 vCPU |
| RAM | 4 GB RAM | 8 GB RAM |
| OS | Node.js 22 LTS or 24 LTS; Linux (Docker or VPS) or Windows with IIS as reverse proxy; SQL Server 2019+ | Node.js 22 LTS or 24 LTS; Linux (Docker or VPS) or Windows with IIS as reverse proxy; SQL Server 2019+ |
| Est. monthly hosting | $25 | $25 |
```bash cp .env.example .env # set DATABASE_URL, MSSQL_SA_PASSWORD, NEXT_PUBLIC_APP_URL docker compose up -d sqlserver docker build -t assistly-ai . docker build --target migrate -t assistly-ai-migrate . docker run --rm --env-file .env assistly-ai-migrate docker run -d -p 3000:3000 --env-file .env assistly-ai ```
```bash npm ci npm run db:deploy npm run db:super-admin npm run build npm start ``` Place the app behind HTTPS (nginx config is in deploy/nginx.conf). Set TRUSTED_PROXY_HOPS=1 in .env.
Quality report
Manually reviewedHow to read the quality report
What this is. Every project on Vibe96 goes through an automated audit of its source code before a reviewer in Dubai checks it by hand. The score is out of 100; 75 is the minimum to be listed, and any single failing check blocks a listing regardless of the score.
- Security
- No real credentials in the code, every data-changing route protected, admin areas role-gated, no SQL built from user input.
- Code structure
- A developer can navigate and extend it: clear module boundaries, no giant files, little copy-paste.
- Docs & completeness
- The setup guide, example config, schema and required files the builder committed to are actually in the archive.
- Dependency health
- Packages pinned, no known critical vulnerabilities, no abandoned libraries in production paths.
- Build & deploy
- A clean machine can restore, build and start it by following the shipped instructions.
Warnings are problems that didn't block the listing; they stay visible here until a later version clears them. The audit reads the code — it does not run it — so always try the live demo before you buy. Lines of code are counted from the same audited archive.
83/ 100
Automated audit score
✓ Passed the auditReviewer notes (4 minor)
Well-organised layered codebase; some duplication and oversized files, plus shipped demo/landing duplicates; mostly reproducible; digest gaps, duplicated docs/landing folders, and a stale script reference; lockfile present, but package.json contents were not provided so versions could not be verified; database rules documented well, but core sql and schema files were not available to verify; authz design looks solid on paper, but no route or action source was visible to verify guards. None of these stop the software from working; the developer has been told and improvements come in a future update.
Versions & updates
No updates yet. New versions appear here with a changelog.
Buyers receive updates free for 1 year.
What you get
- </>Full source codeThe complete audited source, delivered as a download from your account. No obfuscation.
- ▤DocumentationSetup guide, architecture notes, CLAUDE.md and AGENTS.md.
- ↻One year of updatesEvery new version the builder publishes, free for 12 months.
- ♧30 days of supportInstallation and defect support through Vibe96 tickets.