Skip to main content
Developers connect to Grid by running baseten grid setup, which walks them through the rest. Send them a short announcement that names their team, models, and limit. On machines you manage centrally, script setup instead.

​
Before you announce

  • Each team has at least one route, and its harness defaults and spend limits are set.
  • Developers are members of the right teams in Baseten, through invitations or SCIM.
  • Developers can install the Baseten CLI, or you install it for them.

​
Pilot with one team

Before a wider rollout, ask a few developers to connect using the team’s defaults and complete representative tasks. Check model access, MCP tools, usage attribution, and the process for requesting a higher limit. Include developers with custom agent settings so you can review the settings setup replaces. Use the team’s measured usage to choose models and limits. A lower token price alone doesn’t establish lower cost for the same task; compare completed work and usage over the same period. Treat external-provider costs in Grid as estimates.

​
Announcement template

Copy this message into your internal docs, email, or chat, and replace the bracketed parts.

​
What setup changes

Tell developers what to expect before they run setup:
  • Model and endpoint settings are replaced. Setup overwrites the coding agent’s model list, default models, and Baseten endpoint settings without saving the previous values. Developers who pointed their agent at another gateway or proxy lose that configuration.
  • Other settings stay. Setup doesn’t change the MCP servers, plugins, permissions, or other settings in the agent’s configuration files, so MCP servers configured there, including an MCP manager’s, keep working. Claude connectors added through a Claude account don’t work while Claude Code points at Grid; configure those tools as MCP servers in Claude Code instead.
  • Codex signs out of OpenAI. For Codex, setup signs out of OpenAI and ChatGPT and disables ChatGPT login while Grid is configured.
  • Each machine gets its own key. Setup creates a Grid API key for the machine on its first run and reuses it afterward.
To remove the Grid integration settings, a developer runs baseten grid teardown. See Remove Grid from a coding agent.

​
Script setup

baseten grid setup prompts for confirmation, so pass --yes and name the harness and team when you run it from a script or a device-management tool:
Prefer the standard setup so each developer creates their own Grid API key. Spend is attributed to the key’s creator, and personal Baseten API keys aren’t subject to Grid spend limits. Sharing a key across developers loses per-developer attribution. To install an existing key, set BASETEN_API_KEY and pass --use-session-credential. Confirm its owner and scope before distributing it. The CLI never creates, rotates, or deletes that key:

​
Keep agents up to date

When you add, remove, or change a team’s routes or defaults, developers’ coding agents pick up the change the next time setup runs. baseten grid sync replays each machine’s last setup with the team’s current routes, and does nothing when there’s nothing to change, so it’s safe to run on a schedule:
For Codex, sync can restart the background server and interrupt an in-progress turn. With --yes, it skips the confirmation prompt. Schedule sync when developers aren’t actively using Codex.
A route the team has removed falls back to its default, with a warning.