Feature request: expose knowledge base reranking_config (bge-reranker-v2-m3) in the SDK

[HesselTjeerdsma]

Summary

The DigitalOcean Gradient REST API supports configuring a reranker model on a knowledge base via a reranking_config object on the create/update knowledge base endpoints, but the current Python SDK (3.12.1, generated from the Stainless OpenAPI spec) does not expose this field. As a result, SDK users have to drop down to raw HTTP calls to enable reranking, which defeats the purpose of using the typed SDK.

Please regenerate / update the SDK so reranking_config is available on the knowledge base create and update methods.

What exists in the REST API today

The official digitalocean/openapi spec already documents reranking_config in the gen-ai knowledge base examples:

• https://github.com/digitalocean/openapi/blob/main/specification/resources/gen-ai/examples/curl/genai_create_knowledge_base.yml
• https://github.com/digitalocean/openapi/blob/main/specification/resources/gen-ai/examples/curl/genai_update_knowledge_base.yml

Example payload from genai_create_knowledge_base.yml:

{
  "name": "kb-api-create",
  "embedding_model_uuid": "05700391-7aa8-11ef-bf8f-4e013e2ddde4",
  "project_id": "37455431-84bd-4fa2-94cf-e8486f8f8c5e",
  "database_id": "abf1055a-745d-4c24-a1db-1959ea819264",
  "region": "tor1",
  "reranking_config": {
    "enabled": true,
    "model": "bge-reranker-v2-m3"
  }
}

This matches the reranking feature described in the DigitalOcean blog post "Smarter Knowledge Bases for Smarter AI Agents", which announced support for the BAAI bge-reranker-v2-m3 reranker.

What is missing in the SDK

In gradient 3.12.1 (and on main at time of writing):

• grep -ri rerank across the SDK source returns zero matches (verified via GitHub code search: q=rerank repo:digitalocean/gradient-python → 0 results).
• gradient.types.api_retrieval_method.APIRetrievalMethod only contains:
  • RETRIEVAL_METHOD_UNKNOWN
  • RETRIEVAL_METHOD_REWRITE
  • RETRIEVAL_METHOD_STEP_BACK
  • RETRIEVAL_METHOD_SUB_QUERIES
  • RETRIEVAL_METHOD_NONE
• The knowledge_bases.create(...) / knowledge_bases.update(...) methods do not accept a reranking_config parameter, and there is no RerankingConfig / KnowledgeBaseRerankingConfig model type exported.

Proposed change

1. Add a reranking_config parameter to:
   • client.gen_ai.knowledge_bases.create(...)
   • client.gen_ai.knowledge_bases.update(...)
2. Add a corresponding typed model, e.g. KnowledgeBaseRerankingConfig, with fields:
   • enabled: bool
   • model: Literal["bge-reranker-v2-m3", ...] (or str if more models will be added)
3. Expose reranking_config on the response model returned by knowledge_bases.retrieve(...) so callers can read the current state.
4. If reranking ever becomes configurable per-agent (rather than per-KB), expose it there too — but per the current spec it lives on the KB.

Since the SDK is generated by Stainless from the OpenAPI spec, this likely requires updating the gen-ai schema in digitalocean/openapi to declare reranking_config as a real schema (not just an example), and then regenerating.

Why this matters

Without SDK support, every user who wants to enable reranking has to:

• Construct a raw httpx/requests call,
• Manage auth headers separately from the SDK client,
• Forgo type checking and IDE autocompletion on a feature DO is actively marketing.

This is a small, additive change that significantly improves the SDK's coverage of the Gradient AI Platform.

Environment

• SDK: gradient 3.12.1 (latest on PyPI as of 2026-03-25)
• Python: 3.12
• API: https://api.digitalocean.com/v2/gen-ai/knowledge_bases

Thanks!