Answers, with receipts.
We ground every agent response in your real documents — pricing pages, support articles, EHRs, PDFs, internal wikis. With citations, freshness, and answers constrained to what your sources actually say.
A voice agent without
knowledge is a liability.
An agent that invents a refund policy isn't quirky — it's a regulatory finding waiting to happen. Our knowledge layer sits behind your managed AI voice agents so "make something up" is never an option.
Auditable answers
Every grounded answer carries citations. Auditors get the exact chunk and the source URL.
Always fresh
Docs re-sync on a schedule or webhook. The agent always speaks from the latest version.
Refusal that's safe
If retrieval misses, the agent says so out loud and offers to escalate — never improvises.
From PDF to phone call,
in four steps.
Convexa runs a domain-tuned retrieval pipeline for you. You hand us the docs; we build, tune, and maintain everything behind the answers.
We sync your sources
We connect Notion, Google Drive, Confluence, S3, Zendesk, or a custom URL list, then pull your files and watch them for changes.
22 connectorsSemantic chunking
Documents split by topic, not by token count. Headings, tables, and bullet lists keep their structure intact.
~ 360 tokens / chunkHybrid search
BM25 + vector + reranker, in that order. Picks the 3–5 chunks most likely to answer the caller's exact phrasing.
p95 · 80 msConstrained generation
The voice model can only cite from retrieved chunks. Off-source claims are blocked at runtime, not after.
Source-constrainedIf your team writes it,
Convexa can read it.
Docs & wikis
Notion, Confluence, Google Docs, Coda, GitBook, Slite.
File stores
Google Drive, OneDrive, Dropbox, Box, S3. PDFs, DOCX, slides.
Helpdesks
Zendesk Guide, Intercom Articles, Freshdesk, HubSpot KB.
Public web
Sitemap-driven crawls of your pricing page, FAQ, blog. Respects robots.txt.
Structured data
Postgres, MySQL, Snowflake, Airtable. Schema-aware retrieval over rows + columns.
Custom sources
Where your content lives somewhere bespoke, we can connect it as part of setup — embeddings, metadata, and freshness handled for you.
Always-on visibility into
what the agent knows.
We report on what's indexed, when it last synced, and which sources the agent drew on — so you always know what knowledge is behind its answers. It's the same grounding that powers our AI customer support calls.
Where you go next.
We stand up your knowledge base in days — and keep it current for you.