Source: http://www.poma-ai.com/docs/grill/getting-started/

# Getting started

Grill is POMA's end-to-end context engine: it runs the PrimeCut ingestion pipeline (parse, chunk, embed) and persists the result in a managed retrieval layer, so you ingest documents into a project once and then query it. Search is hybrid (dense plus BM25) and the response is prompt-ready context in XML or Markdown rather than raw ranked hits. You run no vector store, reranker or chunking pipeline of your own.

## Before you start

- A POMA account with access to the [Console](https://console.poma-ai.com/). A free account includes 1,000 trial pages with no card required, which covers everything on these pages.
- A **project API key** (prefix `poma_prod_gr_…`), which the SDK reads from `POMA_GRILL_API_KEY`. Account keys (`poma_acc_…`) serve PrimeCut ingestion and cannot call `/grill/*`; each Grill project has its own key.
- Python 3.10+ with `pip install poma`, or any HTTP client if you prefer the raw v3 API.

## The three steps

1. **[Authenticate](/grill/getting-started/authentication)** — generate the project key in the Console and put it in your environment.
2. **[Create a project](/grill/getting-started/projects)** — a project is a namespace for one corpus, optionally assigned to an organisation; ingest and search calls are scoped to it.
3. **[Run the quickstart](/grill/getting-started/quickstart)** — ingest one document, wait for the job to finish, run a hybrid search, and drop the returned context into an LLM prompt. Four lines with the SDK.

After that, the [concepts section](/grill/concepts/) explains ingestion, retrieval and its tiers, the returned context format, and document management, and the [API reference](/grill/reference/api) lists every endpoint and field.