Identify which system owns the run
A model API request and a Spinrun agent task are different interfaces. A model client may call the tool gateway while maintaining its own conversation. The Spinrun web app manages its agent work through its own runtime.
Follow the current client guide
Use the developer documentation for a supported client and endpoint. Do not assume a drop-in OpenAI or Anthropic chat-completions endpoint exists for every agent operation.
Keep credentials and scope separate
Model credentials and connected-tool authorization have different purposes. Configure them through the supported settings and client flow. A model key never grants access to a workspace's external accounts.
Review the result and failures
Validate the output your application needs and handle errors explicitly. For tool writes, verify the external result before retrying a request that may have reached the provider.
Open the relevant screen
Use Model clients and agent runs in the active workspace. Sign in if needed. Check the workspace name before changing shared settings, connections, or billing.
If a control is unavailable, ask the workspace administrator or contact support. Include the screen, the action you tried, and the approximate time. Keep credentials and private customer material out of the request.