[Feature] Support setting max batch size in custom providers

[coandco]

Problem

I'm using huggingface-text-embedding-inference to serve an embedding model. It has a max batch size of 64, and occasionally opencode-codebase-index goes slightly over, leading to the following errors in the hftei logs:

hftei-gina-1   | {"timestamp":"2026-03-29T00:38:18.185898Z","level":"ERROR","message":"batch size 66 > maximum allowed batch size 64","target":"text_embeddings_router::http::server","filename":"router/src/http/server.rs","line_number":1233,"span":{"name":"openai_embed"},"spans":[{"name":"openai_embed"}]}
hftei-gina-1   | {"timestamp":"2026-03-29T00:38:18.296007Z","level":"ERROR","message":"batch size 68 > maximum allowed batch size 64","target":"text_embeddings_router::http::server","filename":"router/src/http/server.rs","line_number":1233,"span":{"name":"openai_embed"},"spans":[{"name":"openai_embed"}]}
hftei-gina-1   | {"timestamp":"2026-03-29T00:38:19.015925Z","level":"ERROR","message":"batch size 68 > maximum allowed batch size 64","target":"text_embeddings_router::http::server","filename":"router/src/http/server.rs","line_number":1233,"span":{"name":"openai_embed"},"spans":[{"name":"openai_embed"}]}

Would it be possible to add a max batch size to the custom provider definition so I can fully index my codebase without hitting the limit?

Proposed Solution

I'd like a max_batch_size configuration parameter in the custom provider configuration that limits how many requests are sent as a batch.

Alternatives Considered

N/A

Additional Context

N/A

[Helweg]

Implemented this on branch \ and opened PR #41: #41

This adds support, accepts as a config alias, and splits custom OpenAI-compatible embedding requests into smaller calls for strict servers like text-embeddings-inference.

Verification run locally:

• npx vitest run tests/custom-provider.test.ts
• npx vitest run tests/config.test.ts
• npx vitest run tests/embeddings.test.ts
• npm run typecheck
• npm run build
• npm run lint

[Helweg]

Follow-up with clean formatting: implemented on branch feat/custom-provider-max-batch-size and opened PR #41: #41

This adds customProvider.maxBatchSize support, accepts max_batch_size as a config alias, and splits custom OpenAI-compatible embedding requests into smaller /embeddings calls for strict servers like text-embeddings-inference.

Verification run locally:

• npx vitest run tests/custom-provider.test.ts
• npx vitest run tests/config.test.ts
• npx vitest run tests/embeddings.test.ts
• npm run typecheck
• npm run build
• npm run lint