fix: use native Ollama API endpoints instead of OpenAI-compatible routes #7071
Conversation
![roomote[bot]](https://avatars.githubusercontent.com/in/1546624?s=60&v=4){kind=small}
- Replace OpenAI client with direct axios calls to Ollama native API - Use /api/chat endpoint instead of /v1 OpenAI-compatible endpoint - Update tests to reflect the new native API implementation - Maintain backward compatibility with existing configurations Fixes #7070
dosubot
bot
added
size:XL
![ellipsis-dev[bot]](https://avatars.githubusercontent.com/in/64358?s=60&v=4){kind=small}
| }) as const, | ||
| ) | ||
|
|
||
| let buffer = "" |
There was a problem hiding this comment.
Unused variable 'buffer' declared here isn't used; please remove for clarity.
| let buffer = "" |
| for (const matcherChunk of matcher.update(delta.content)) { | ||
| yield matcherChunk | ||
|
|
||
| // Convert Anthropic messages to Ollama format |
There was a problem hiding this comment.
The inline conversion of Anthropic messages to Ollama format (lines 82–117) could be extracted into a utility function for improved readability and maintainability.
This comment was generated because it violated a code review rule: irule_tTqpIuNs8DV0QFGj.
| }) as const, | ||
| ) | ||
|
|
||
| let buffer = "" |
There was a problem hiding this comment.
Is this buffer variable intentional? It's declared but never used. Could we remove it to keep the code clean?
| const ollamaMessages: OllamaMessage[] = [{ role: "system", content: systemPrompt }] | ||
|
|
||
| // Convert messages to Ollama format | ||
| for (const message of messages) { |
There was a problem hiding this comment.
I notice we check for deepseek-r1 models and adjust temperature, but unlike the previous OpenAI implementation, we don't apply R1 format conversion to the messages. Is this intentional? DeepSeek R1 models might need special message formatting for optimal performance.
| } else { | ||
| throw new Error(`Ollama completion error: ${error.message}`) | ||
| } | ||
| } | ||
| if (error instanceof Error) { |
There was a problem hiding this comment.
This error handling seems redundant - axios errors are already handled above. Could we remove this duplicate handler to simplify the error flow?
| model: string | ||
| created_at: string | ||
| message: { | ||
| role: string |
There was a problem hiding this comment.
Consider using the more specific type "system" | "user" | "assistant" instead of string for better type safety?
| apiKey: "ollama", | ||
| timeout: getApiRequestTimeout(), | ||
| }) | ||
| this.baseUrl = this.options.ollamaBaseUrl || "http://localhost:11434" |
There was a problem hiding this comment.
The default Ollama URL appears in multiple places. Would it be cleaner to extract this to a constant like const DEFAULT_OLLAMA_BASE_URL = "http://localhost:11434"?
hannesrudolph
added
the
Issue/PR - Triage
This PR fixes issue #7070 where Ollama models were incorrectly using OpenAI-compatible routes instead of native Ollama API endpoints.
Problem
When using models like
gpt-oss:120bwith Ollama, the plugin was trying to get completions from OpenAI routes (/v1) instead of the native Ollama/api/chatendpoint.Solution
/api/chatendpoint for chat completions instead of/v1OpenAI-compatible endpointChanges Made
Modified
src/api/providers/ollama.ts:/api/chatendpoint using axiosUpdated tests:
src/api/providers/__tests__/ollama.spec.ts: Updated to mock axios instead of OpenAI clientsrc/api/providers/__tests__/ollama-timeout.spec.ts: Updated timeout tests to work with new implementationTesting
Fixes #7070
Important
This PR updates
OllamaHandlerto use native Ollama API endpoints with axios, replacing OpenAI-compatible routes, and updates tests to reflect these changes.OllamaHandler./api/chatendpoint for chat completions.ollama.ts.ollama.spec.tsandollama-timeout.spec.tsto mock axios and test new implementation.This description was created by
for 14c33f8. You can customize this summary. It will automatically update as commits are pushed.