# The new search

July 7, 2026 · New feature · https://vamotalent.com/docs/changelog/agent-search

> Research agents source developers from your company's context, qualify the best candidates, and learn from your feedback.

Search Github has a new engine. Describe the role in plain language, and a research agent reads that description along with any context you attach, sources developers by the work they have actually shipped on GitHub, and strengthens each match with LinkedIn. Candidates arrive in rounds of 50, and a thumbs up or down on any card shapes the next round.

## Sourcing runs in rounds

Every search returns a round of 50 developers. The agent reads your request and any context you added, finds people through the projects they build, then confirms the fit with their LinkedIn profile. A round is a complete set to work through, and every round is new people, never a repeat of a name you have already seen.

When a round is thin, the agent widens on its own before it gives up: it broadens the technical interests and fans out to similar companies, and it never relaxes the location you set. When you want more, **Source another round** produces a fresh 50 tuned by everything you have rated so far.

## The filters, grouped the way a search actually works

Open **Refine search** in the top right to see the plan the agent drafted, organized into three tabs.

**Context** holds what the search was built from: the **Location** (the one hard requirement), your **Original request**, an attached **Job description**, and any **Context links** like a company site or a blog post.

![The Context tab with Original request, Job description, Context links, and Location](https://vamotalent.com/docs/changelog/agent-search/context.png)

**GitHub** is what their code should be about:

- **Subjects**: the domains the work covers, like dating apps or physics simulation.
- **Skills and technologies**: languages, frameworks, and tools.
- **Projects**: seed repositories to search near, set to **Similar** or **Exact** contributors.
- **Project stars** and **Followers**: ranges that shape ranking.
- **Early career signals**: surfaces learning-in-public work, courses, and up-and-comers.
- **AI focused**: biases every angle toward agents, MCP, LLMs, and inference.

![The GitHub tab with Subjects, Skills, Projects, Stars, and Early career and AI signals](https://vamotalent.com/docs/changelog/agent-search/github.png)

**LinkedIn** is who they are professionally:

- **Require a LinkedIn match**: narrows to people with a linked profile.
- **Experience**: Early-career, Up to senior, or All levels.
- **Unconventional paths**: product managers, solutions engineers, and second-career builders who ship real code.
- **Individual Contributors**: engineers at any level including staff and principal, without founders, executives, and people-managers.
- **Likely Looking**: a soft read on who may be open to a move.
- **Companies**: match on current employer or anyone who has worked there, with a **Current** and **All** switch.
- **Current company size** and **School**.

Every filter is a strong signal that shapes ranking. Location is the only hard requirement, so a round always fills instead of coming back empty.

![The LinkedIn tab with experience, role, company, and education filters](https://vamotalent.com/docs/changelog/agent-search/filters.png)

## Rate a card to steer the next round

Each result carries the evidence for why it matched: the strength signals and confirmed LinkedIn details under **Who they are**, and the specific proof from its GitHub work under **Why they match**. A thumbs up or thumbs down on any card feeds the agent. Up reinforces what it found, down lets you say what ruled the person out, and the next round reflects both.

![A result card showing Who they are and Why they match with a thumbs up and down](https://vamotalent.com/docs/changelog/agent-search/card.png)

## How to use it

**1. Sign in and open Search Github**

Sign in at [vamotalent.ai/login](https://vamotalent.ai/login). In the left sidebar, open **Search Github**.

**2. Describe the role**

Type the role in plain language and add any context: a job description, a company link, a location. The agent drafts a search plan and starts the first round.

**3. Work the round**

Review the round of 50. Thumbs up the strong ones, thumbs down the misses and note why. Open a card's LinkedIn and GitHub links to dig in.

**4. Refine or source more**

Open **Refine search** to adjust any filter, then **Save and Source New Round**. Or click **Source another round** for a fresh 50 tuned by your ratings.

## What to expect

Rounds land in under a minute and carry a repo on every card, so the reason a person surfaced is on screen. Ratings compound: the more you react, the closer each round gets to the role. Earlier searches stay in the sidebar under Search Github, so you can return to any of them and pick up where you left off.

> **Tell us when a result is off**
>
> Every screen has a **Feedback** link. If a card looks wrong or a round comes up thin, send a note and it reaches the team directly.