> ## Documentation Index
> Fetch the complete documentation index at: https://internal.september.wtf/llms.txt
> Use this file to discover all available pages before exploring further.

# Prerequisites

> What you need installed before running the Engine locally.

The Engine runs inside Docker, so the host requirements are minimal. You
need:

## Required

* **Docker** with Compose v2. Versions tested:
  * Docker Desktop 4.x (macOS, Windows).
  * Docker CE 24+ (Linux).
* **Git** for cloning the repo.
* **An LLM provider API key**. Pick one:
  * Anthropic — `sk-ant-...`
  * OpenAI — `sk-...`
  * Google Gemini — Google API key with the Generative Language API enabled.
* **An OpenAI API key for embeddings.** This is required regardless of
  which chat provider you choose. The Engine uses OpenAI's
  `text-embedding-3-small` for memory search.

## Optional

* **`sqlite3` CLI** if you want to inspect the brain database from the host.
* **`jq`** for pretty-printing the JSON streams in your terminal.
* **An MCP server** (Slack, Gmail, etc.) if you want to test external
  connectors. The Engine itself does not require any.

## Disk and memory

* Disk: the Engine image is \~1.2 GB. Brain databases start at a few MB and
  grow with use; budget 1 GB per active user as a comfortable upper bound
  for a single-user development setup.
* RAM: the Engine process is light (\~200 MB resident). The bottleneck is
  whatever the LLM call returns.

## Architecture

Tested on:

* macOS arm64 (Apple Silicon).
* Linux x86\_64.

If you're on Apple Silicon, `docker-compose.yml` already pins the test
container to `linux/amd64` because some sandbox tests require x86 syscall
behavior. You don't need to do anything special — it just works.

## Network

Outbound HTTPS to your chosen LLM provider must be reachable. If you're
behind a corporate proxy:

* Set `HTTPS_PROXY` and `HTTP_PROXY` in your shell before running
  `docker compose up`.
* Make sure those envs are passed into the container by the compose file
  (the default compose file does this).

You should not need any inbound port forwarding for local development —
everything is served on `localhost:8000`.

## Confirm everything is in place

```bash theme={null}
docker --version
docker compose version
git --version
echo "$LLM_API_KEY" | head -c 10  # should print the first 10 chars of your key
echo "$OPENAI_API_KEY" | head -c 10
```

If those all return reasonable output, head over to
[Quickstart](./quickstart).