GTM Lead Agent

An AI pipeline for inbound leads: research each company with Exa, score it against your ICP with Claude, log everything to a Google Sheet, and drop a personalized outreach draft in Gmail.
Built on One. Nothing is ever sent automatically.

Website  ·  Docs  ·  Dashboard

---

What it does

Give it a lead (or a CSV of leads) and for each one it:

1. Researches the company with Exa: what they do, recent news, funding, launches.
2. Scores the lead 1-10 against your ICP (you describe it in plain English in config/icp.md) using Claude, with a written rationale and talking points.
3. Logs the row to a Google Sheet your whole team can see: score, rationale, talking points, sources.
4. Drafts the outreach email in your Gmail drafts folder, personalized with a concrete detail from the research. It never sends; you review and hit send.

All three platform calls (Exa, Sheets, Gmail) go through One, so there is exactly one auth setup and one SDK instead of three API integrations.

$ npm start -- leads.example.csv

Processing 3 leads

── jane@ramp.com
   researched ramp.com: 5 sources
   fit 7/10: Ramp is a fintech building AI agents into their product...
   gmail draft created (r880107...)
   logged to sheet

Setup (about 5 minutes)

1. Connect your accounts through One

npm i -g @withone/cli
one init          # create an account or log in
one add exa
one add google-sheets
one add gmail

2. Configure this repo

git clone https://github.com/withoneai/gtm-lead-agent
cd gtm-lead-agent
npm install
cp .env.example .env

Fill in .env:

• ONE_SECRET: your One API key (in ~/.one/config.json after one init, or the dashboard)
• ANTHROPIC_API_KEY: from platform.claude.com
• The three connection keys: run one --agent list and copy the key for exa, google-sheets, and gmail

3. Create your lead log sheet

npm run setup

This creates a spreadsheet with the right headers and prints the LEADS_SHEET_ID line to paste into .env.

4. Describe your ICP

Edit config/icp.md. It ships with an example; replace it with your product, your ideal customer, and your email voice. The scoring model reads this file verbatim, so plain English is exactly right.

Usage

# a CSV with an email column (name, company, notes optional)
npm start -- leads.csv

# a single lead
npm start -- --email jane@acme.com --name Jane --notes "asked about pricing"

# see the assessment and draft without writing anything
npm start -- --email jane@acme.com --dry-run

Behavior worth knowing:

• Drafts are only created for leads scoring 5+. Everything is logged to the sheet either way.
• Leads on free email providers (gmail.com etc.) with no company name skip research and cap at 4/10.
• --dry-run prints the score and the drafted email to the terminal and writes nothing.

How it's built

Step	Platform	How
Research	Exa	@withone/sdk typed action exa.search
Scoring + drafting	Claude (claude-opus-5)	@anthropic-ai/sdk structured outputs (Zod schema)
Lead log	Google Sheets	@withone/sdk googleSheets.appendValuesSpreadsheetRange
Outreach draft	Gmail	@withone/sdk gmail.createUsersDraft

~400 lines of TypeScript in src/. The One SDK handles auth, request signing, and retries for every platform call; swapping Sheets for HubSpot or adding Slack notifications is a few lines because the SDK types 30+ platforms the same way.

Extending it

• Different CRM: @withone/sdk ships typed actions for HubSpot, Airtable, Notion, and more. Replace src/sheet.ts.
• Trigger on form submits: wire your form webhook to run npm start -- --email ....
• Slack alerts for hot leads: after scoring, post 8+ leads to a channel via the One Slack actions.

License

MIT