Gretel Logo

Introduction: Client

The Gretel Python Client allows you to interact with the Gretel REST API. Where possible, we have added specific conveinence to help auto-manage things like ingest of large amounts of records. Please see the getting started section and the module documentation for more details.

In order to use this client, you must have a valid API key. Please login to our console in order to get one.


For basic installation:

$ pip install gretel-client

If you are working with Pandas and wish to enable the features for working with DataFrames:

$ pip install gretel-client[pandas]

Getting Started

The easiest way to get started is by creating a Project instance that allows you to directly interface with a Gretel project.

Gretel will automatically manage API calls based on the project instance you are working with.

First, you will need to create a Client instance:

from gretel_client import get_cloud_client

client = get_cloud_client('api', 'your_api_key')

Next, you can create a Project instance. This high-level object let’s you interact direclty with your project. To do this, you can use the get_project factory method that is part of the Client instance.


Get an existing project

You can get an existing project by:

project = client.get_project(name='my_existing_project')

This will raise a BadRequest error if the project does not exist or you are not a member.

Create a project

You can create a new named project by using the create flag.

project = client.get_project(name='my_named_project', create=True)

If the project exists or the name is available, a Project instance is returned. Otherwise a BadRequest is raised.

Optionally, if a name does not matter, you can use an auto-named project and Gretel will create one for you:

project = client.get_project(create=True)

You can see the name Gretel created by checking

Flushing and Deleting projects

Once you have a project instance you can flush and delete it.

Assuming an instance variable name of project, you can:


This will purge all data from the metastore and delete cached annotated records. Your project namespace, permissions, and collaborators will still be there.



This will flush all data (see above) and also delete the project shell (name, description), collaborators, and permissions.


NOTE: Both of these commands run asyncronously in Gretel as they are called. So it may be a few moments for them to complete. Additionally, your project instance, if using delete() will not be usable anymore.

Sending Records

The Gretel API consumes JSON-formatted records. There’s a few ways that you can send these via the client.


You can send Python dictionaries directly using different methods.

First, you can send them on an API-call-by-API-call basis. When records are received by the Gretel API, each record is assigned a gretel_id and the status of each record (success or fail) is returned to the user. You can utilize this behavior with the send() method:

This method will return a tuple of success and failure notifications.

s, f = project.send({'foo': 'bar'})

The list of success items will look like:

[{'idx': 0, 'gretel_id': '8afbd2c8f5b147c9bd30faffa3e5ad0a'}]

idx refers to the index in the original list of records that were sent. If you only sent a single record (a dict) then idx 0 will refer to that record.

The failure list will be empty if there were no issues on ingest.

You may also send a list of dicts. When using send() you will be subject to the max record size per API call, which is currently 50.

data = [{'foo': 'bar'}, 1, 2, 3]
s, f = project.send(data)

Here only the first record is valid:

[{'idx': 0, 'gretel_id': '4c34d2fbc2974f81af2c07ac78eedcef'}]

And the last three were not valid JSON objects:

[{'idx': 1,
  'message': 'Individual records must be JSON objects',
  'sender_fault': True},
 {'idx': 2,
  'message': 'Individual records must be JSON objects',
  'sender_fault': True},
 {'idx': 3,
  'message': 'Individual records must be JSON objects',
  'sender_fault': True}]

Bulk Records

If you are streaming data or sending large amounts of records, you may not want to worry about using the send() method and chunking up your records. For this use case, you may use the send_bulk() method. It has the same signature as send() but the client will automatically chunk the input up and multithread to push the records to Gretel.

One exception to using this is that you will not receive any success or failure confirmations.

data = [{f'foo_{i}': 'bar'} for i in range(500)]

# >> 500 records [00:00, 157184.23records/s]

Module Reference: Client