Feature · Platform

Model accounts

Manage model accounts, credentials, budgets and spend.

Last changed

Generated from commit be57775 ·

How it works

What you can connect

  • API keys: Anthropic, OpenAI, Google Gemini and OpenRouter.
  • Amazon Bedrock with an IAM access key (Claude models only).
  • Google Vertex AI with a service account key.
  • Azure OpenAI with a key and the resource endpoint.

You can have one account of each kind for yourself, and the organization can have one of each kind.

What never connects

A Claude or ChatGPT sign-in, its tokens or its login files. Agent refuses them, and admin keys and temporary keys too. To use a Claude or ChatGPT plan, see Subscriptions and vendor terms.

Connect an account

  1. Open Model accounts and select Connect.
  2. Choose the provider and paste the credential.
  3. Optional: choose the tiers it serves, and a daily and monthly budget.
  4. Select Check and connect. Agent makes one free, read-only call. When the check fails, nothing is stored.

Only owners and admins can connect organization accounts. Any member can connect their own.

The order

For a call, Agent tries your own accounts first (as the person who asked for the work), then the organization's, then the platform's. An assignee's key never pays for work someone else asked for. When an account fails or its budget is spent, the call goes to the next one.

Fall back to the platform (on by default) lets the platform's models answer when your accounts cannot. When it is off, the call fails instead. A tier that no account serves always runs on the platform.

Statuses

Works (ready), Refused (invalid) and Unreachable: the check could not reach the provider. Select Test to check again. A key that the provider refuses during work stops being used until it is tested again.

How keys are stored

Encrypted in the vault. The page shows only the last four characters, or the service account's email. Disconnect deletes the credential and keeps its past spend.

From the help article Model accounts and API keys, checked against the product on .

What you can do

The operations this feature's screens call. Each one is POST /api/<name>, and the badge says who may call it.

  • listModelAccounts

    ViewerNo API tokens

    Lists model accounts.

    Types: EmptyRequest returns ListModelAccountsResponse

  • disconnectModelAccount

    MemberNo API tokens

    Disconnects the model account.

    Types: ModelAccountIdRequest returns OkResponse

  • updateModelAccount

    MemberNo API tokens

    Updates the model account.

    Types: UpdateModelAccountRequest returns ModelAccountDto

  • testModelAccount

    MemberNo API tokens

    Tests the model account.

    Types: ModelAccountIdRequest returns ModelAccountDto

  • updateModelAccountSettings

    AdminNo API tokens

    Updates model account settings.

    Types: UpdateModelAccountSettingsRequest returns ModelAccountSettingsDto

  • startModelAccountConnect

    MemberNo API tokens

    Starts the model account connect.

    Types: StartModelAccountConnectRequest returns StartModelAccountConnectResponse

  • connectModelAccount

    MemberNo API tokens

    Connects the model account.

    Types: ConnectModelAccountRequest returns ModelAccountDto

  • listMembers

    Viewer

    Lists members.

    Types: EmptyRequest returns ListMembersResponse

Screens

What it records

The feature writes these events to the audit log, the permanent record of what happened.

  • Integration changed
  • Integration connect failed
  • Integration connect started
  • Integration connected
  • Policy changed

Generated from code at commit be57775 ·

Turn the numbers into shipped work.

Agent runs these choices for you: a persistent AI worker with memory and rules, on your Claude and Codex subscriptions.