> ## Documentation Index
> Fetch the complete documentation index at: https://docs.trygravity.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# How it works
> The request flow, the auction, and what makes matching contextual.
## The request flow
```mermaid theme={null}
sequenceDiagram
participant Client as Your client
(browser / app)
participant Server as Your server
participant LLM as LLM provider
participant Gravity as Gravity Engine
Client->>Server: messages + gravity_context
par
Server->>LLM: chat completion
LLM-->>Server: streamed tokens
and
Server->>Gravity: POST /api/v1/ad
Gravity-->>Server: matched ad
end
Server-->>Client: SSE stream (tokens + ads)
```
1. The user sends a message. Your client fires the LLM request **and** calls `gravityContext()` to capture device + session info.
2. Your server receives both and fires `gravity.getAds()` in parallel with the LLM call. The ad request never blocks streaming.
3. Gravity's engine matches the conversation against active campaigns, runs an auction, and returns the winning ad.
4. Your stream appends the ad as a final chunk. Your client shows it.
Everything happens in a few hundred milliseconds end-to-end. Since it runs in parallel with your LLM call, user-perceived latency doesn't move.
## Contextual matching
When a user asks *"how do I deploy a serverless Postgres database?"*, the engine looks at the recent conversation and returns an ad whose campaign targets that topic — say, a managed Postgres provider. The ad copy is generated on the fly to fit the conversation, so the user sees something relevant and specific rather than a generic banner.
There's no persistent user profile driving this match. It lives and dies with the conversation.
## What gets paid out
You get paid when the ad becomes visible to the user. CPM is the public billing model on Gravity today.
## What's next
Build the integration end-to-end.
Glossary of the terms used throughout the docs.