Set up Grid for your organization

Set up Grid once per team: add its members, choose the models it can use, and set its spend limits. When you finish, developers on the team can run baseten grid setup and see the team’s models in their coding agent. Manage Grid from the Grid page in the Baseten app. Use Activity and Users to review usage. Choose a team on Routes, Connections, or Guardrails to manage its models, provider keys, or budgets. You can also manage routes, provider connections, and team settings with the Baseten CLI, or manage every setting with the Management API.

Before you begin

Baseten has enabled Grid for your organization. To request access, Contact us.
You’re an organization Admin, or an Admin of each team you set up. Team members can use a team’s routes but can’t change them.
To use Anthropic, OpenAI, or xAI models, you have an API key from that provider. See Choose a provider API key.

For the CLI procedures, install the Baseten CLI and sign in to your organization.

Credentials and ownership

Grid uses different credentials for administration, coding agents, and upstream model providers:

Credential	Who uses it	Purpose
Baseten sign-in	Admins and developers	Sign in to the Baseten app or CLI with your organization account.
Grid API key	A developer’s configured coding agents	Authenticate model requests to Grid. Grid attributes spend to the key’s creator and applies limits in the team. Standard setup creates or reuses this key.
Provider API key	An admin connects it to a team	Allow the team’s routes to call Anthropic, OpenAI, or xAI. The provider bills the account behind the key. Developers don’t need to install it on their machines.

Use an authorized Baseten API key for Management API operations. It is separate from the inference-only provider key you add in Connections.

Organize teams and members

Grid uses your organization’s teams. Each team has its own routes, provider connections, harness defaults, and spend limits. Create a team for each group of developers that needs different models or a different budget, such as an infrastructure team with frontier models and a $10,000 team-wide limit, and a go-to-market team with open-weight models only. Invite developers to a team, or sync team membership from your identity provider with SCIM. A developer on several teams picks one with --team when they run setup.

Connect model providers

Skip this step if the team uses only Baseten Model APIs. Connecting a provider supplies its credential; you still need to add its models to the team.

UI
Baseten CLI

To connect a model provider:

Sign in to your workspace at app.baseten.co and choose Grid in the sidebar.
Choose Connections.
Choose the team from the dropdown.
Choose Connect for the provider.
Enter a provider API key restricted to inference, then save. See Choose a provider API key.

Confirm that the provider appears as connected for the team.To replace an existing key, choose the provider’s edit icon.

To connect a model provider:Replace <team> with your team’s name or ID. This example connects Anthropic. For OpenAI or xAI, change the provider and use a distinct secret name, such as openai-key or xai-key, in both commands. Reusing a secret name replaces its value and affects connections that already use it.

Save the provider key as a team secret. Run the following command and enter the key at the prompt:
baseten org secret set --team "<team>" --name anthropic-key

Create a connection that uses that secret:

baseten grid connection create --team "<team>" \
  --provider anthropic --secret-name anthropic-key

Check that the connection lists the intended provider and secret:
baseten grid connection list --team "<team>"

For more information, see baseten grid connection create.

Baseten stores the key as a secret on the team and never shows it again. The provider bills its own usage to that key. To rotate the key, edit the connection; every route that uses the connection switches to the new key.

Choose a provider API key

Give Grid a key that can run inference but can’t administer your provider account. A service account key isn’t automatically restricted to inference. Check both the account’s permissions and the key’s scope before you add it.

Provider	Use	Don’t use
Anthropic	A workspace-scoped service account key without organization administration permissions	An Admin API key
OpenAI	A restricted service account key that can only perform inference	An admin key or an unrestricted key
xAI	An inference API key with access to the required endpoints and models	A Management API key

For key creation and permissions, see the Anthropic authentication guide, OpenAI project API key guide, and xAI API key permissions.

Choose team models

Add the models a team’s developers can pick in their coding agent. Mix open-weight models from Model APIs with frontier models from a connected provider, so developers can use a frontier model for the hardest problems and an open-weight model for routine work without leaving their agent.

UI
Baseten CLI

To add models to a team:

Sign in to your workspace at app.baseten.co and choose Grid in the sidebar.
Choose Routes.
Choose the team from the dropdown.
Choose Add routes.
Choose Model APIs or a connected provider.
Add each model that the team should be able to call.

Confirm that the new models appear in the team’s routes.

To add models to a team:Replace <team> with your team’s name or ID. For a Model API, replace <model> with its model name and run the following command:

baseten grid route create --team "<team>" \
  --target-type baseten-model-api --target-model "<model>"

For a model from a connected provider:

List the team’s connections and copy the connection’s ID:
baseten grid connection list --team "<team>"

Replace <provider-model> with the provider’s model name and <connection-id> with the ID from the previous step:

baseten grid route create --team "<team>" \
  --target-type connection --target-model "<provider-model>" \
  --target-connection-id "<connection-id>"

For either model type, confirm that the new model appears in the team’s routes:

baseten grid route list --team "<team>"

For more information, see baseten grid route create.

Grid lists a model in a coding agent only when the agent can call the model’s API and the model supports tool calling. See Model requirements.

Add an auto route

An auto route lets a classifier choose among several direct routes for each request. The default route handles a request when the classifier doesn’t choose one. Auto routes can’t contain other auto routes. Add the direct routes the classifier can choose before creating the auto route.

UI
Baseten CLI

To add an auto route:

Sign in to your workspace at app.baseten.co and choose Grid in the sidebar.
Choose Routes.
Choose the team from the dropdown.
Choose Add routes.
Choose Intelligent routing.
Enter the route’s display name.
Choose the allowed routes and a default route from that set.
Choose an available classifier.
Choose Add intelligent route.

Confirm that the auto route appears in the team’s routes with the intended classifier and allowed models.

To add an auto route:Replace <team> with your team’s name or ID.

List the team’s direct routes and copy the IDs of the routes to include:
baseten grid route list --team "<team>"
List available classifiers and copy the classifier’s ID:
baseten grid classifier list

Create the auto route. Replace the placeholders with the IDs from the previous steps. The default must be one of the allowed routes:

baseten grid route create --team "<team>" \
  --target-type classifier --classifier-id "<classifier-id>" \
  --allowed-route-id "<primary-route-id>" \
  --allowed-route-id "<background-route-id>" \
  --default-route-id "<primary-route-id>" \
  --display-name "Coding assistant"

Replace <auto-route-id> with the new route’s ID and check its classifier, allowed routes, and default route:
baseten grid route describe --id "<auto-route-id>"

For more information, see baseten grid route create and baseten grid classifier list.

Developers can select the auto route in their coding agent like any other compatible team route.

Set harness defaults

Set harness defaults to choose the model each coding agent uses when a developer runs setup without picking one. Claude Code and OpenCode have a primary model and a background model for lightweight tasks; Codex and Pi have a primary model only. To pair a frontier model with a cheaper one, set a frontier route as the primary model and an open-weight Model API as the background model. Developers get the frontier model for their main work, while lightweight background requests go to the open-weight model. Without an override, Baseten recommends routes from the team’s LLM Model APIs. It recommends the highest-ranked Model API for primary work and the least expensive Model API for background work. Connected-provider routes aren’t automatic recommendations, but you can select a compatible provider route as an override.

UI
Baseten CLI
Management API

To choose a coding agent’s default models:

Sign in to your workspace at app.baseten.co and choose Grid in the sidebar.
Choose Routes.
Choose the team from the dropdown.
Under Harness model overrides, expand Harness overrides.
For Default, choose the primary route for the coding agent.
For Background, where available, choose the route for lightweight tasks.
Choose Save changes.

Confirm that the selected routes remain in the team’s harness overrides.

To choose a coding agent’s default models:Replace <team> with your team’s name or ID. This example sets Claude Code’s primary and background routes.

List the team’s routes and copy the IDs of the primary and background models:
baseten grid route list --team "<team>"

Save those route IDs as Claude Code’s defaults:

baseten grid team-settings update --team "<team>" \
  --harness claude-code \
  --primary-route-id "<primary-route-id>" \
  --background-route-id "<background-route-id>"

Check that Claude Code lists the intended primary and background routes:

baseten grid harness-defaults --team "<team>" --harness claude-code

For more information, see baseten grid team-settings update and baseten grid harness-defaults.

To choose a coding agent’s default models:Set BASETEN_API_KEY to a Baseten API key with team Admin permissions. Find the team’s ID with List teams and replace <team_id> with it. This example sets Claude Code’s primary and background routes.

List the team’s routes and copy the IDs of the primary and background models:

curl --get "https://api.baseten.co/v1/grid/routes" \
  --header "Authorization: Bearer $BASETEN_API_KEY" \
  --data-urlencode "team_id=<team_id>"

Save those route IDs as Claude Code’s defaults:

curl --request PATCH "https://api.baseten.co/v1/grid/team_settings" \
  --header "Authorization: Bearer $BASETEN_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{"team_id":"<team_id>","harness_defaults":{"claude_code":{"primary":"<primary-route-id>","background":"<background-route-id>"}}}'

Read the team’s settings to confirm the saved routes:
curl --get "https://api.baseten.co/v1/grid/team_settings" \ --header "Authorization: Bearer $BASETEN_API_KEY" \ --data-urlencode "team_id=<team_id>"
Check harness_defaults.claude_code.primary.route.id and harness_defaults.claude_code.background.route.id against the IDs you saved.

For more information, see Update team settings and Get team settings.

Only the defaults you provide change. Developers who use team defaults pick up the changes when they next run setup or sync. Explicit route choices take precedence over team defaults.

Set spend limits

Set a per-member limit for the team, a team-wide limit, or both. See Set spend limits and view usage for how limits combine and how to change one developer’s limit.

Verify the team is ready

Before inviting the whole team to connect, have one member complete developer setup and send a test request. Confirm that:

The intended team models appear in the coding agent’s model picker.
The selected model returns a response.
Existing MCP tools still work if the team relies on them.
Usage appears for that developer after the metering delay, which can be up to 15 minutes.

Roll out to developers

Once the pilot works, use Roll out Grid to developers for an announcement template and scripted setup options.

Get started

Model APIs

Baseten Grid

Inference

Frontier Gateway

Training

Sandboxes

Development

Deployment

Engines

Model Labs

Organization

Hosting options

Observability

Troubleshooting

Set up Grid for your organization

Before you begin

Credentials and ownership

Organize teams and members

Connect model providers

Choose a provider API key

Choose team models

Add an auto route

Set harness defaults

Set spend limits

Verify the team is ready

Roll out to developers

​Before you begin

​Credentials and ownership

​Organize teams and members

​Connect model providers

​Choose a provider API key

​Choose team models

​Add an auto route

​Set harness defaults

​Set spend limits

​Verify the team is ready

​Roll out to developers

Before you begin

Credentials and ownership

Organize teams and members

Connect model providers

Choose a provider API key

Choose team models

Add an auto route

Set harness defaults

Set spend limits

Verify the team is ready

Roll out to developers