Features
An assistant you can let talk to customers.
Most assistants can produce a plausible answer. The hard part is guaranteeing it is true — and that is where your customers' trust is won or lost.
Grounded in your documents
The assistant only answers from your documents, and suggests the right YouTube video when there is one. The passage it used remains available in your console.
It stays silent when it doesn't know
Below a measured relevance threshold no model is called at all: the assistant returns your fallback message instead of an invention.
On your site in five minutes
One snippet to paste, and the widget takes on your colours. Also available through the API for your own channels.
Your data stays yours
Four successive isolation barriers separate tenants, right down to the database row.
Four steps, one hour.
Drag the steps: the diagram follows.
A question? Start here.
Still stuck?
+224 625 56 73 58A conversational assistant you train on your own documents: guides, contracts, catalogues, FAQs. It answers your customers from that corpus and nothing else. Drop it on your site with one snippet, or call it through the API from your own channels.
An hour is enough for a genuinely useful first assistant: time to upload your documents, wait for indexing, set the welcome message and paste the snippet. Fine-tuning — tone, suggested questions, forbidden topics — comes later, without taking the service down.
PDF, Word, Excel, PowerPoint, plain text and Markdown, up to 50 MB per file. You can also index a website by its address, a question-and-answer list typed by hand, or a YouTube transcript. Scanned PDFs go through automatic character recognition.
Every question is matched against your corpus before any language model is called. If the retrieved passages fall below a measured relevance threshold, no model is called at all: the assistant returns the fallback message you wrote. This is the product's central safeguard, and it is on by default.
It answers in the language of the question, whatever language your documents are in. French is the default for the interface and for answers; English is supported end to end.
One script tag before the closing body tag. The widget picks up your colours, logo and welcome message from the console, with no further technical work. For bespoke channels — mobile app, phone system, WhatsApp — the REST API is documented.
Priced on real usage, not on promises.
Every plan includes hybrid search, answers grounded in your documents and the relevance threshold. What changes is the volume.
Discovery
To test the product on your real documents.
- 100 messages per month
- 1 assistant
- 10 indexed documents
- 1 user
- History kept for 7 days
- Widget shows “Powered by Sutura Chat”
Essential
For a first customer-facing rollout.
- 1,500 messages per month
- 1 assistant
- 100 indexed documents
- 2 users
- History kept for 90 days
- Widget shows “Powered by Sutura Chat”
Professional
For a support desk running every day.
- 6,000 messages per month
- 3 assistants
- 500 indexed documents
- 5 users
- History kept for 365 days
- Widget without our branding
- Webhooks and conversation export
Business
For several brands, several teams, sustained volume.
- 25,000 messages per month
- 10 assistants
- 3,000 indexed documents
- 20 users
- History kept for 730 days
- Widget without our branding
- Webhooks and conversation export
- Bring your own model keys (BYOK)
Enterprise
Negotiated volumes, dedicated hosting, a contractual uptime commitment and tailored support.
Your assistant can be live this afternoon.
Upload your documents, set the tone, paste one snippet. The trial needs no credit card.