AI customer support for SaaS

AI customer support for SaaS companies: account, plan, billing and onboarding questions answered from your docs, seat math done in code, proration handed to your team, plus API and webhooks.

In short: OctaDezx is AI customer support for SaaS companies. It answers account, plan, billing, SSO setup and onboarding questions from your own documentation, computes seat totals and plan price differences in code when your plans are saved as priced items, and hands prorated charges, credits and account changes to your team because those depend on the customer's billing record. It runs on your website and in your product through the chat widget or chat link, on Slack, Discord, email and messaging apps, and connects to your backend through a REST API, signed webhooks and an MCP server.

Software customers ask how a feature works, why a login failed, what a plan includes, what twenty seats would cost and why this month's invoice looks different. OctaDezx answers the questions your docs and price list already settle, does the arithmetic in code instead of in a language model, and routes the questions that need an account record to the person who owns it.

The everyday problem

SaaS support queues fill with how-to and plan questions the docs already answer, while billing disputes and incidents wait behind them. An AI that guesses a prorated amount or promises a plan change it cannot make turns a support question into a billing problem.

How OctaDezx helps

Import your docs and plan details, put the agent on your site, in your app and on your community channels, and connect your helpdesk. Questions with verifiable answers are answered from your sources. Questions that need the customer's subscription record, such as proration or a credit, are handed to your team with the conversation and the reason attached.

SaaS support questions it handles

A compound billing question

Illustrative flow, not a customer transcript. It assumes per-seat plan prices are saved as items in OctaDezx.

Plan, look up, calculate, verify

Single questions take the fast path. A message that asks several things at once is split into a plan of steps, and each step runs with only the inputs it needs: a lookup in your saved plans and policies, money math in code, and a final verifier that refuses any amount, promise or action no verified fact supports. If a step lacks evidence, the assistant asks a targeted question or hands off rather than filling the gap.

Every written reply then passes the brand safety check against your verified facts, and corrections your team makes become proposed lessons that only an owner can approve. The method and the current holdout safety measures are published.

Connect it to your product and backend

  • Chat widget and chat link. Add the website widget with one script tag, or open the shareable chat link from your product's help menu. This is the in-app path available today.
  • Headless Agent API. Start and continue conversations from your server at https://api.octadezx.com/v1, with statuses, handoff reasons, filed actions and optional Server-Sent Events streaming. Keep API keys server side.
  • Signed webhooks. Receive HMAC-signed events for escalations, orders, appointments, voicemail and refund requests, and act on them in your own systems.
  • MCP server. Let your support and success team read escalations, reply in thread and update knowledge from Claude, ChatGPT, Gemini, Cursor or your own agent.
  • Node.js SDK. npm install @octadezx/sdk wraps the Agent API with streaming, signs user tokens and verifies webhooks. Version 0.1.0, MIT licensed.
  • In-app messenger. npm install @octadezx/messenger puts the chat in your product with React bindings, and signed user tokens let the agent treat a signed-in customer as that customer. A Python SDK is in early access.

What it does not reach

Frequently asked questions

Explore the connected tools

Read more on this

OctaDezx home · Pricing · Blog · Start free trial