fix(indexing): pass configured dimension to OpenAI-compatible embedder

[magiccodelab]

Fixes #14579

Summary

Pass options: { dimensions: config.modelDimension } when instantiating OpenAICompatibleEmbedder in CodeIndexServiceFactory.createEmbedder().

Why

When configuring an OpenAI-compatible embedding provider with a custom dimension (such as 4096 for MRL-enabled models like Qwen3-Embedding-8B or OpenAI text-embedding-3), the vector store table was created with the configured dimension, but OpenAICompatibleEmbedder never received config.modelDimension.

As a result, HTTP requests to the embedding endpoint did not include the "dimensions" parameter, causing the provider to fall back to its server default dimension (e.g. 768) and triggering a dimension mismatch error during semantic_search:
GenericFailure, Invalid input, No vector column found to match with the query vector dimension: 768

This change brings openai-compatible in line with the other providers (openrouter, ollama, kilo) which already forward config.modelDimension.

Testing

• Added unit tests in packages/kilo-indexing/test/kilocode/indexing/service-factory.test.ts verifying that dimensions is forwarded to OpenAI-compatible embedding requests when configured, and omitted when unset.
• Ran package unit tests and typechecks (bun test, tsgo --noEmit).

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

The change correctly forwards config.modelDimension into OpenAICompatibleEmbedder (positional args match the constructor: baseUrl, apiKey, modelId, maxItemTokens, options), consistent with kilo, ollama, and openrouter. The embedder already conditionally spreads dimensions, so an unset value stays omitted, and both the SDK and full-endpoint request paths pick it up. resolveEmbeddingProfile yields no static dimension for openai-compatible, so the vector store already required an explicit dimension; this closes the mismatch. Tests cover the set and unset cases.

Files Reviewed (3 files)

• .changeset/openai-compatible-embedding-dimensions.md
• packages/kilo-indexing/src/indexing/service-factory.ts
• packages/kilo-indexing/test/kilocode/indexing/service-factory.test.ts

---

_{Reviewed by deepseek-v4.1-flash · Input: 0 · Output: 0 · Cached: 0}

_{Review guidance: REVIEW.md from base branch main}