Embeddings Feature is very rigid #5572
Description
App Version
3.26
API Provider
Not Applicable / Other
Model Used
Not listed
Roo Code Task Links (Optional)
No response
🔁 Steps to Reproduce
💥 Outcome Summary
Problem Description:
Users are encountering embeddings:validation.configurationError when configuring codebase indexing with "OpenAI Compatible" providers (e.g., DeepInfra, as shown in the screenshot). The current "OpenAI Compatible" integration appears too rigid, failing due to minor variations in API request/response formats that differ from a strict OpenAI specification.
Proposed Solution:
Implement a flexible, programmable mechanism for integrating embedding providers. This should allow users to:
Define Custom HTTP Requests: Specify the request (method, headers, body) using a template or curl-like syntax. or just give a fetch command...
Program Response Parsing: Provide logic (e.g., JavaScript snippet, JSONPath) to extract embeddings from arbitrary API response structures (e.g., resp.data[0].embeddings).
Motivation:
This enhancement will enable universal compatibility, supporting a wide range of services like DeepInfra, Voyage AI, and custom hosted models, without requiring specific, hardcoded integrations. It empowers users to configure the system for any RESTful embedding API.
Impact & Current Workaround:
Until this is implemented, users are limited to strictly OpenAI-compliant services or must wait for official, specific integrations for their desired providers.
Context:
Screenshot: See attached original image for the error and current setup.
Example API: https://deepinfra.com/Qwen/Qwen3-Embedding-4B/api