# Gemini API

> How larch uses the Gemini API: Scout’s AI sales agent scores leads from 1 to 5, drafts outreach and reads company web pages with Google’s Gemini models.

URL: https://larch.ee/technologies/gemini-api/

larch builds AI features on Google’s Gemini models. Scout’s sales agent uses the Gemini API to score leads against the ideal customer profile, draft personalized outreach and pull facts out of web pages.

- Products built: 1
- Default model: Gemini Flash
- Models to pick from: Flash-Lite · Flash · Pro
- Integration: REST, no SDK
- Lead score: 1 to 5

## Built with Gemini API

One agent, on Gemini. Scout’s agent does the reading and writing a salesperson would do, and the Gemini API is the model behind it.

### [Scout](https://larch.ee/work/scout/)

Scout calls the Gemini API wherever the work needs judgement: scoring each lead from 1 to 5, drafting outreach from the user’s knowledge base, pulling contacts out of web pages and summarizing documents.

- Year: 2025
- Stack: Next.js, React, TypeScript, Tailwind CSS, SQLite, Playwright, AI agents, Gemini API

## Related service

### [AI and automation](https://larch.ee/services/ai-automation/)

Useful intelligence, wired into the tools your team already works in.

What you get:

- **AI agents that work in production, not just demos**: Scout’s agent runs 10+ cycles on Gemini to reach a lead target.
- **MCP servers, LLM orchestration and multi-API pipelines**: Scout chains Gemini calls with web search and page scraping in one pipeline.
- **Scraping, data enrichment and automated daily workflows**: Gemini reads the pages Playwright opens and pulls out names, roles and companies.

## How we use Gemini API

A model for the judgement calls. Most of Scout’s pipeline is plain code. Gemini handles the parts that need judgement: is this company a fit, who should we write to, and what should the email say.

### Scout: where Gemini comes in

Scout’s agent works through a pipeline: discover companies, enrich contacts, qualify, expand and draft. Most of the steps are plain code. The Gemini API comes in where the work needs judgement.

Qualification is the first. Every lead gets a score from 1 to 5 against the user’s ideal customer profile, on industry, company stage, team size, budget, tech stack and founder type, and the model explains its reasoning.

Drafting is the second. Outreach emails are written from the user’s knowledge base of case studies, pricing documents and company profiles, so each draft speaks to that company rather than to everyone.

Reading is the third. When Playwright opens a company’s website, Gemini reads the page and pulls out the people, roles and details the next step needs. It also summarizes the documents in the knowledge base, so the drafts can draw on them.

### A thin integration

Scout calls the Gemini REST API directly, with no SDK in between. The default model is Gemini Flash, and in Settings the user can pick another Gemini model, from the lighter Flash-Lite to Pro, to trade speed for depth. Because the integration is this thin, a new model is a setting, not a rewrite.

### Moving to Gemini

Scout first reached its models through a gateway that hid the provider. In July 2026 it moved to the Gemini API directly. The pipeline stayed the same; only the calls behind it changed.

## Other technologies

Used alongside it. Scout’s agent runs on Gemini, and our MCP servers bring tools into AI assistants. See what else we built with them.

- [AI agents](https://larch.ee/technologies/ai-agents/): The agent that runs Scout’s whole B2B sales pipeline, and Bridddge’s AI automation.
- [MCP servers](https://larch.ee/technologies/mcp-servers/): seo-mcp brings SEO audits into AI editors, and Bridddge ships with its own MCP server.
- [All technologies](https://larch.ee/technologies/): Eleven technologies, each one running in a product we shipped.

## How to start

Tell us what you want to ship. Bring the idea. We’ll tell you how we’d build it and what it takes.

- Book a 30-minute call: https://calendly.com/nicholas-stackwise/30min
- Email: nicholas@larch.ee