> For the complete documentation index, see [llms.txt](https://docs.eyepop.ai/developer-documentation/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.eyepop.ai/developer-documentation/sdks/python/configuration.md).

# Configuration

Credentials and environment variables

The SDK reads your API key from `EYEPOP_API_KEY`. Create one in the [dashboard](https://dashboard.eyepop.ai).

```shell
export EYEPOP_API_KEY=eyp_...
```

You can pass it directly instead, though the environment variable keeps it out of your source:

```python
endpoint = EyePopSdk.sync_worker(api_key="eyp_...")
```

If your backend already holds a short-lived token, pass it as `access_token=` instead of an API key. A token targets `https://api.eyepop.ai`, the legacy worker API. The Compute API requires an API key — pointing `eyepop_url` at it with only a token raises `Compute API endpoint (https://compute.eyepop.ai) requires EYEPOP_API_KEY` — so `session_uuid` and `EYEPOP_SESSION_UUID` are ignored on the token path, and a persistent Deployment needs `api_key`.

### Optional variables

| Variable              | Description                                                                                                                                                                                                                                                                  |
| --------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `EYEPOP_SECRET_KEY`   | Secret key authentication, which works with named Pops. `EYEPOP_API_KEY` does not, and has to be unset — the SDK reads both, and a named Pop with an API key present raises `EYEPOP_API_KEY can only be used with transient pops`.                                           |
| `EYEPOP_POP_ID`       | Named Pop to run. Defaults to `transient`.                                                                                                                                                                                                                                   |
| `EYEPOP_SESSION_UUID` | Attach to a persistent Deployment instead of creating a transient session.                                                                                                                                                                                                   |
| `EYEPOP_ACCOUNT_ID`   | Required for some Data API calls.                                                                                                                                                                                                                                            |
| `EYEPOP_URL`          | Override the API base URL.                                                                                                                                                                                                                                                   |
| `EYEPOP_LOG_LEVEL`    | Log verbosity for the `eyepop` logger tree. Unset, the SDK installs no handler of its own and its records fall through to your application's logging configuration. `eyepop.requests` stays at `WARNING` unless the level is `DEBUG`; `LOG_LEVEL` is accepted as a fallback. |

### Transient and persistent sessions

With no session UUID and a `pop` — the flow every example here uses — the SDK creates a new **transient** session each time it connects, and deletes the pipeline it created on exit. It does not delete the session itself. This is the right default for building and testing.

With no `pop`, the SDK reuses an existing non-persistent session when one is available, and leaves both it and its pipeline untouched on exit.

To run against a persistent Deployment, pass its session UUID. The Pop is fixed when the Deployment is created, so you do not pass one:

```python
with EyePopSdk.sync_worker(session_uuid="<your-session-uuid>") as endpoint:
    result = endpoint.upload("photo.jpg").predict()
```

### Local mode

An [on-premise instance](https://docs.eyepop.ai/deploying/on-premise) serves the EyePop runtime on your own machine. **Local mode** points the SDK at `http://127.0.0.1:8080` instead of the cloud and sends no account credentials — the instance is already registered to your account, and reaching the loopback port is what authorizes the client.

```python
from eyepop import EyePopSdk
from eyepop.worker.worker_types import InferenceComponent, Pop

pop = Pop(components=[
    InferenceComponent(ability="eyepop.person:latest")
])

with EyePopSdk.sync_worker(is_local_mode=True, pop=pop) as endpoint:
    result = endpoint.upload("image.jpg").predict()
    print(result)
```

`EYEPOP_LOCAL_MODE=true` in the environment selects it without the argument. Pass `eyepop_url` for an instance on another port or host:

```python
endpoint = EyePopSdk.sync_worker(
    is_local_mode=True,
    eyepop_url="http://127.0.0.1:9090",
    pop=pop,
)
```

Connecting creates a pipeline on the instance and disconnecting removes it, so keep the `with` block and reuse one connected endpoint for many images.

### Next steps

* [Running Inference](/developer-documentation/sdks/python/inference.md) — process files, streams, URLs, and video
* [Composable Pops](/developer-documentation/sdks/python/composable-pops.md) — chain models into a pipeline
* [On-Premise](https://docs.eyepop.ai/deploying/on-premise) — create an instance to run local mode against