Skip to content

Add DeepSeek V3/R1 provider to the Inference effect #450

@aallan

Description

@aallan

One-row addition to the _PROVIDERS registry introduced in #413. DeepSeek exposes an OpenAI-compatible endpoint so no custom response parsing is required.

DeepSeek V3

  • Base URL: https://api.deepseek.com/v1/chat/completions
  • Auth: Authorization: Bearer {key} (OpenAI-style)
  • Default model: deepseek-chat
  • Response path: data["choices"][0]["message"]["content"]
  • Env var: VERA_DEEPSEEK_API_KEY

DeepSeek R1 (reasoning model)

  • Same endpoint, model name deepseek-reasoner
  • Response path identical; reasoning trace in reasoning_content field (ignored)

Both models share the same API key and base URL. Recommend deepseek-chat as the default model (faster, cheaper for typical Vera workloads) with R1 selectable via VERA_INFERENCE_MODEL.

Implementation
After #413 lands, this is a single entry added to the _PROVIDERS registry in vera/codegen/api.py, plus an env-var fetch in host_inference_complete and a documentation update in spec/09-standard-library.md.

Depends on #413.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or requestintegrationIntegration with external tools and languages

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions