Research agent · 2026

A research agent for sales reps.

I built GTM Intelligence because account research is the kind of work that takes 30 minutes and produces a paragraph. Reps have to do it before every discovery call. Most either skip it or do it badly. I wanted to see if a research agent could do the boring part well enough that the rep just walks into the call ready.

Open in new tab ↗
01

The output is a paragraph.

Most sales tooling dumps everything about a company into a dashboard. Reps don't have time to read a dashboard before a call. They want to know what changed recently and what to lead with.

GTM pulls funding, product launches, leadership moves, and recent press from the past quarter. The model writes a one-paragraph briefing. If there is nothing meaningful to report on an account, the briefing says so instead of inventing things.

02

Every claim has a source.

The model isn't allowed to write anything that doesn't trace back to a document it actually fetched. Each sentence in the briefing has a citation underneath it that opens the original article, filing, or post.

If the model can't find a source for something, it leaves that part out. A shorter briefing that's right beats a longer one that's confidently wrong.

03

Runs overnight against your calendar.

Add your week's calls on Sunday night. The briefings run overnight and arrive in your inbox by Monday morning. Updates fire only when an account has new material since the last briefing, so quiet weeks stay quiet.

What it runs on

A short tech sheet.

Model
Claude Sonnet 4.6
Retrieval
Hybrid (BM25 + dense)
Sources
News, LinkedIn, EDGAR, Crunchbase
Latency
~6s per briefing
Cost
~$0.04 per briefing
Cache
Source-layer, 1h TTL
Stack
FastAPI, SSE, Postgres
Hosted on
Render, Cloudflare Workers
← back All things next → PR Diff Review