Back to all workCase study, AI sales agent

GNX Sales

An AI sales agent that finds buyers, researches why they would care, writes personal outreach and reaches them by email or AI voice call.

TypeSaaS product
Our roleDesign, build, AI, infra
StatusLive at gnxsales.com
GNX Sales homepage

The problem

Outbound sales is mostly hours of prep: finding the right companies, working out why each one might care, writing a message that does not read like a template, then following up. GNX Sales hands that work to an agent, so the team only steps in for conversations that are worth having.

What the agent does

1

Find

Sources leads that match the ideal customer.

2

Research

Works out why each buyer might care.

3

Write

Drafts outreach grounded in that research.

4

Reach

Sends email or places an AI voice call.

5

Book

Puts interested buyers on the calendar.

The hard part:not making things up

An agent that writes to real buyers cannot invent facts about them. We built a context layer that keeps three kinds of knowledge apart, and only verified facts are allowed into a message.

Verified facts

Checked against a source. Safe to use in outreach.

Hypotheses

Likely, but unconfirmed. Used to plan, never stated as fact.

Unknowns

Gaps the agent knows it has, so it asks instead of guessing.

Each run also saves a hashed snapshot of the context it used, so any message can be traced back to exactly what the agent knew when it wrote it.

22klines of TypeScript
123API endpoints
40backend services
9background job workers

Built with

Next.jsNode.js + ExpressTypeScriptOpenAI Agents SDKAzure OpenAIRetell voice AIApolloRedis + BullMQSupabase PostgresPython research workerGoogle Workspace APIsSentryPostHog

Want an agent like this for your team?