Skip to main content
AI setup starts with large provider cards for ChatGPT, Claude, and Local / custom. After choosing a provider, select a model from the dropdown; one to five yellow dots show relative token usage. You can also enter an exact custom model ID. The dots are a comparison guide, not an exact price or context-window measurement. OpenAI and Anthropic connections accept an API key. OpenAI-compatible connections also require a base URL, with HTTPS required except for loopback HTTP endpoints such as Ollama on localhost. API keys are moved into the operating-system credential vault rather than stored in workflow JSON. Gmail setup uses OAuth PKCE. Discord and Slack setup accept incoming webhook URLs and store those secrets in the same host-owned vault.

​
Manage a connection

  1. Choose Add connection, select the provider, and complete its setup.
  2. Give the connection a name that identifies its real account or destination.
  3. Use Test to verify current access.
  4. Select the named reference in a node.
  5. Review affected workflows before renaming or revoking.
The status can be connected, expired, revoked, error, or setup required. Last-used and expiry metadata help diagnose access without revealing the credential. Revocation blocks future use immediately. An exported or synced workflow keeps only a requirement/reference and must be reconnected in its destination environment. AI connections can power the AI workflow builder, the AI node, and Write with AI in the Code editor. Prompts and relevant workflow or code context are sent to the selected provider; the API key is not. Bug reporting is a first-party sndbox service and does not use a connection from this screen. See Report an app bug.