Hostname for calling service #526
Conversation
shershkop-ironclad
left a comment
shershkop-ironclad
left a comment
There was a problem hiding this comment.
Review comments from local validation.
| @@ -1008,7 +1007,8 @@ export const ChatNodeBase = { | |||
| const isMultiResponse = data.useNumberOfChoicesInput || (data.numberOfChoices ?? 1) > 1; | |||
|
|
|||
There was a problem hiding this comment.
[P0] Restore the default OpenAI URL when openAiEndpoint is blank. openAiEndpoint is normalized to '' by the processor/app settings, so ?? never falls back here. With the default configuration (no custom endpoint set), this now builds /chat/completions instead of https://api.openai.com/v1/chat/completions, and the same pattern in the new assistant/file/thread calls yields relative URLs as well. That breaks ordinary OpenAI usage for anyone who leaves the endpoint empty.
| @@ -1008,7 +1007,8 @@ export const ChatNodeBase = { | |||
| const isMultiResponse = data.useNumberOfChoicesInput || (data.numberOfChoices ?? 1) > 1; | |||
|
|
|||
| // Resolve to final endpoint if configured in ProcessContext | |||
There was a problem hiding this comment.
[P1] Keep openAiEndpoint as a full chat endpoint here. The existing setting, docs, and presets all treat openAiEndpoint as the complete chat completions URL, for example LM Studio's http://localhost:1234/v1/chat/completions or Azure deployment URLs with ?api-version=.... Appending /chat/completions turns those into invalid URLs like .../chat/completions/chat/completions or appends the path after the query string, so custom chat endpoints stop working.
| } | ||
|
|
||
| const response = await fetch(`https://api.openai.com/v1/assistants/${assistantId}`, { | ||
| const baseUrl = context.settings.openAiEndpoint ?? 'https://api.openai.com/v1'; |
There was a problem hiding this comment.
[P2] Do not reuse the chat endpoint as the assistants/files base URL. This is only safe when openAiEndpoint is an API root, but the setting is currently exposed as a Chat-node endpoint and is commonly configured to a full /chat/completions URL. In that case assistant/thread/file nodes now request paths like .../chat/completions/assistants/... instead of OpenAI's API, so enabling a custom chat endpoint breaks these nodes even if they should still talk to the default OpenAI service.
1 participant
Fix to not hardcode hostname for service and also allow embedding to take argument