Skip to main content
Do not use this method for new deployments. We recommend using our Terraform self-hosting instead.
This guide walks through the process of setting up Braintrust in your AWS account. Before following these instructions, make sure you have account access and are able to sign into Braintrust. There are two methods you can follow:
  1. Use the braintrust command-line (CLI) tool. This is the recommended method and requires just a single command.
  2. Use the AWS console. If you need more control or cannot get AWS credentials for the CLI, this option allows you to run the whole process within the CLI.

​
Setting up the stack: CLI

1

​
Install the CLI

Install the latest Braintrust CLI:
pip install --upgrade 'braintrust[cli]'

# verify the installation worked
braintrust --help
2

​
Create the CloudFormation stack

There are just a few relevant parameters you should consider for most use cases:
  • A stack name. This is arbitrary and allows you to refer back to the stack later. A name like “braintrust” or “braintrust-dev” should be fine.
  • --org-name should be the name of your organization (you can find this in your URL on the app, e.g. https://www.braintrust.dev/app/<YOUR_ORG_NAME>/...). This will ensure that only users with access to your organization can invoke commands on your Braintrust endpoint.
  • --provisioned-concurrency the number of lambda workers to keep running in memory. This is useful if you expect to have a lot of concurrent users, or if you want to reduce the cold-start latency of your API calls. Each increment costs about $40/month in AWS costs. The default is 0.
  • --template if you are deploying in a region other than us-east-1, you should specify the template for that region.
braintrust install api --create <YOUR_STACK_NAME> \
  --org-name <YOUR_ORG_NAME> \
  --provisioned-concurrency 1
This command is idempotent. If it fails, re-run the command without the --create flag, and it will resume tracking the progress of your CloudFormation stack. If your stack enters a failed state, e.g. it fails to create, please reach out to support.
Once the install completes, you’ll see a log statement like
Stack with name braintrust has been updated with status: UPDATE_COMPLETE_CLEANUP_IN_PROGRESS
Universal URL: https://dfwhllz61x709.cloudfront.net
Save the endpoint URL. You can now skip ahead to the Verifying the stack section.

​
Setting up the stack: CloudFormation console

1

​
Create the CloudFormation

The latest CloudFormation template for Braintrust is available in the following regions:

​
US East 1


​
US East 2


​
US West 2


To start in us-east-1, click here to open up the CloudFormation setup window. If you prefer, you can also click “Create Stack” directly in the CloudFormation console and specify one of the URLs above as the S3 template link.Setup CloudFormation
These instructions walk through how to setup through the AWS UI, but you are welcome to install it via the command line too if you prefer.
You do not need to set any parameters to create a stack; however, there are several you can set to configure behavior. The most important ones to consider while creating a stack are:
  • OrgName - The name of your organization. This will restrict access to users in your organization. By default it is set to * which means all users can query the endpoint. However, rest assured, only users in your org will be able to access its resources (due to access control checks).
  • ProvisionedConcurrency - The number of lambda workers to keep running in memory. This is useful to set if you want to minimize the cold-start latency of your API calls. Each increment costs about $40/month in AWS costs. The default is 0.
Once you fill in the parameters, accept the acknowledgments and select Create stack to start creating the template. This can take a few minutes (up to ~10) to provision the first time.Acknowledgements
Behind the scenes, the template sets up a few key resources:
  • A VPC with public & private subnets for networking
  • A few lambda functions which contain the logic for executing Braintrust commands
  • An API gateway that runs commands against the lambda functions
  • An auto-scaling group for Brainstore
  • A Postgres database, Redis cache, and S3 buckets
Redis Performance Note: If you’re experiencing performance issues, avoid using burstable Redis instances (t-family instances like cache.t4g.micro). These instances use CPU credits that can be exhausted during high-load periods, leading to performance throttling. Consider upgrading to non-burstable instances like cache.r7g.large or cache.r6g.medium for predictable performance.
2

​
Getting the Universal URL

Once the stack is provisioned, you should see UPDATE_COMPLETE as its status:Update completeNavigate into the stack, select the Outputs tab, and copy the value for UniversalURL. We’ll use this to test your endpoint and configure Braintrust to access your org through it. For the rest of the doc, we’ll refer to it as <YOUR_UNIVERSAL_URL>.Universal URL

​
Verifying the stack

Run the following command to test that the stack is running. The first time you run it, AWS may do some setup work to provision your lambda function, and it can take up to 30 seconds to run. 
curl -X GET '<YOUR_UNIVERSAL_URL>/ping'
You should see a response like {"id":"80c48a10-4888-4382-a55b-255018e70fe5","email":"[email protected]","organizations":[]}.

​
Configure your organization’s endpoint

Visit your organization’s settings page and set the API URL captured above. You can skip the realtime URL and proxy URL, unless you have an advanced need that requires it (see the docker guide for more information). Once you configure the URL, select Save. The page automatically attempts to test that you’re authorized to access the URL. URL config
The braintrust install api command tries to install these values for you. So if you see them already filled in, no need to change them!

​
Test the application end-to-end

Hooray! At this point you should be able to test the full application. The easiest way to do this is by using the Python SDK. This simple Python script will run a full loop of using Braintrust and setting up an experiment. 
import braintrust

# NOTE:
# * You can specify your API key in the `api_key` parameter, or set it as an environment variable
#   `BRAINTRUST_API_KEY`. If you don't specify it, the SDK will look for the environment variable.
# * You should not specify api_url in the SDK. It will automatically use the value you set in the
#   settings page.
experiment = braintrust.init(project="SetupTest", org_name="my_org.com", api_key="sk-****")
experiment.log(
    input={"test": 1},
    output="foo",
    expected="bar",
    scores={
        "n": 0.5,
    },
    metadata={
        "id": 1,
    },
)
print(experiment.summarize())

​
Maintaining your installation

Most new Braintrust releases do not require stack updates. Occasionally, however, you will need to update the stack to get access to new features and performance enhancements. Like installation, you can update the stack through either the CLI or AWS console.

​
Using the Braintrust CLI

To update your stack, run the following (replacing <YOUR_STACK_NAME>): 
braintrust install api <YOUR_STACK_NAME> --update-template
You can also use this command to change parameters, with or without template updates. For example, if you want to allocate provisioned concurrency 8 to your lambda functions, run 
braintrust install api <YOUR_STACK_NAME> --provisioned-concurrency 8

​
Using the AWS interface

You can also update the stack directly through AWS. Their docs walk through how to update through the console and the aws cli. The rest of the guide covers topics only needed for advanced configurations.

​
Connect Braintrust’s VPC to other internal resources

To permit incoming and outgoing traffic between Braintrust’s lambda functions and other internal cloud resources, you can either run everything in the same VPC or setup VPC peering. This is necessary if you want to access resources like an LLM gateway or a database that is not publicly accessible.

​
VPC Peering

When you create your Braintrust CloudFormation, it automatically creates a VPC with the same name as your CloudFormation. You can access the Braintrust VPC’s name and ID from the CloudFormation’s Outputs tab (named pubPrivateVPCID). VPC Peering AWS has a comprehensive guide for configuring peering. Follow the instructions for

​
Troubleshooting

  • If you continue to see errors after updating the VPC peering group, you may need to update your CloudFormation template (which will effectively reboot your Lambda functions). You can do this by triggering an update on the CloudFormation and letting it run. You may need to change a stack parameter and then change it back to trigger the updates.
  • You can manually test network settings by booting up an EC2 machine in the Braintrust VPC to test connectivity. Make sure to assign a public IP to the instance and use the public subnet of the VPC while initializing.