Skip to content
Qevra
Now in private development

AI support infrastructure for software companies.

Connect Qevra through an API and resolve customer issues using your own product knowledge. Send the question and the context you have; get back an answer, a confidence score, the sources behind it and a clear signal when a human should step in.

REST · JSON · versioned from v1

POSThttps://api.qevra.dev/v1/resolve
Request
{
  "message": "I can't send WhatsApp messages",
  "user_id": "usr_8f21ac",
  "context": {
    "plan": "growth",
    "channel": "whatsapp"
  }
}
Response
{
  "status": "resolved",
  "confidence": 0.91,
  "answer": "Your WhatsApp sender is pending review…",
  "suggested_actions": ["verify_sender"],
  "sources": ["doc_whatsapp_setup"],
  "escalation_required": false
}

Support that compounds instead of repeating

Most support tooling treats every ticket as new. Qevra treats a resolved ticket as an asset.

Resolve

Send a support request with whatever context you already have. Qevra returns a grounded answer, a confidence score, the sources it used and whether a human should take over.

confidence
0.91
sources
doc_whatsapp_setup
suggested_actions
verify_sender
escalation_required
false

Learn

Every resolution your team approves becomes reusable organisational knowledge. Issues you solved once stop consuming agent time.

  • Approved answers reused
  • Knowledge from real tickets
  • No manual curation

Scale

One versioned API, per-organisation isolation and usage you can actually read. Model providers stay behind an abstraction, so your integration does not change when they do.

  • Multi-tenant by default
  • Usage per organisation
  • Provider-agnostic

One integration point

Qevra sits behind whatever your customers already use to reach you. It does not replace your inbox, your helpdesk or your chat widget.

Your product
  • Support inbox
  • In-app chat
  • Helpdesk webhook
Qevra API
  1. RetrieveYour docs, FAQs and resolved tickets
  2. ReasonGrounded in your product context
  3. ReturnAnswer, confidence, sources, actions
Outcome
  • Resolved automatically
  • Draft for an agent
  • Escalated to a human
Approved resolutions flow back into the knowledge index

It shows its work, and says when it cannot

Every request lands in your workspace with the answer, the confidence it carried, the latency and the documents it stood on. Approve it, or write the answer that should have been given — that correction becomes a document the next question is answered from.

QevraRequests
  • I can't send WhatsApp messages

    Answered

    Your WhatsApp sender is pending review. Meta usually clears this within 24 hours and nothing on your side is misconfigured — you will start sending as soon as it is approved.

    12 Mar 09:41confidence 91.2%412ms1 source: WhatsApp sender setupusr_8f21ac
  • Do you charge for API calls that fail?

    Corrected

    Failed calls are not billed.

    Corrected by a human, and indexed

    Calls that return a 4xx are not billed. A call that returns a resolution is billed even when confidence is low, because the model ran.

    12 Mar 09:12confidence 62.4%388ms2 sources: Billing overview, Pricing FAQ
  • Can you refund the annual plan I bought yesterday?

    No knowledge
    11 Mar 17:03204msusr_3ba07e

Start with a workspace and an API key

Create an organisation, connect your knowledge, and call the API from your own product.