AI chat that knows your offer.
Serves customers 24/7.
Add the chat to your site and upload your pricing, policies and FAQ. Cortify answers customers around the clock, books appointments, checks orders and hands trickier cases to your team.
1
Two pillars
that make the difference.
An assistant that knows your product and can actually do something with it. Everything else is detail.
Answers grounded
in your documents.
You upload PDFs, DOCX files and pages. Cortify indexes them hybridly — vectors (pgvector + HNSW) plus BM25 full-text — and the assistant cites the source with every answer.
Functions that do the work,
not just talk.
Per-tenant functions. The assistant decides when to call them, with what arguments — and shows the result to the user.
Replies in the user's language.
Auto-detection. EN, PL, DE, UK, CS — no configuration.
Quality you can measure, not just claim.
Every deployment is tested with evaluation question suites: an AI judge scores faithfulness to sources, relevancy and correctness after each knowledge-base change. With the Quality Module that dashboard lives in your own panel.
A widget that does
exactly what you want.
Turn each of these on with a single toggle in the panel — per client. No code, no redeploy.
Live responses
Text appears word by word, in real time.
Source citations
A "Sources" section under each answer lists the documents it relied on.
Voice input
Users dictate questions by mic — transcribed server-side.
Read aloud
A speaker icon on each answer reads it out loud.
File uploads
Drop an image, PDF, TXT or Markdown straight into the chat.
Conversation history
Users return to past threads right where they left off.
Answer feedback
Thumbs up / down under every answer — a quality signal.
Sound notifications
A subtle chime when the assistant sends a new message.
Pill launcher
A horizontal bar with avatar and last-message preview instead of a bubble.
Suggested prompts
Ready-made starters so users know what to ask.
Light & dark mode
The widget matches the host page or forces a chosen theme.
Tool calls
Expandable cards show which tool the assistant invoked.
This is a subset of the most-used widget features. The operator configures the full list of toggles, limits and add-ons (semantic cache, analytics, Quality module, and more) in the panel.
Four steps. Every
conversation runs the same.
You upload documents.
PDF, DOCX, pages. Ingestion API or panel. Incremental updates.
It finds the right context.
Hybrid retrieval: vectors (pgvector + HNSW) plus BM25 full-text (VectorChord), fused with RRF. Optional rerank.
Answers, citing the source.
LiteLLM gateway picks the model, fallback chain on failure, per-tenant budget.
Calls functions, if needed.
Tool calling: checks the order, books a slot, escalates to a human.
Write to the assistant.
An interactive preview of real conversation scenarios: knowledge-base answers, source citations and function calls — exactly how the widget behaves on a client's site.
See how the chat answers from your own content.
Enter your website address. Our chat reads a few of its pages and shows the questions your customers ask, with answers grounded in your content.
Takes about a minute. No sign-up, no commitment.
Observability someone actually reads.
Full traces in self-hosted Langfuse, per-tenant cost, prompt versioning per industry, deflection metrics — all in one view.
Customer data stays
where it belongs.
Every company's data kept apart.
One customer's conversations, documents and settings never reach another. The database enforces it on every query, and the widget only works on domains you allow.
Data in the EU. DPA available.
Servers and AI models in the European Union. You choose how long conversations are kept, and export and deletion are in the dashboard.
Docker Compose.
Your hardware, your models.
For Enterprise customers: the full stack on your infrastructure. No outbound calls to external services if you use a local LLM.
$ docker compose up -d → postgres ✓ → keydb ✓ → langfuse ✓ → cortify-api ✓
One-off setup.
Flat monthly fee.
You pay once for setup: teaching the chat your offer, configuration and launch. Then a flat monthly fee with a conversation allowance, with post-launch care included.
Quality Module
A quality dashboard in your own panel: Quality Score 0–100, your own test question suites scored by an AI judge, and an answer review queue.
- Quality Score and quality trends after every knowledge-base change
- Regression tests on your own questions (faithfulness, relevancy, correctness)
- Automatic flagging of suspicious answers + human review
What people usually ask.
20-minute demo.
On your documents.
We'll show you a working assistant on your documentation. If you don't see the value after 20 minutes — you just don't deploy.