/model-deepseek

DeepSeek V4 adapter — thinking: { type, effort } + auto-prefix prompt cache.

Part of wasmagent — a TypeScript + WASM agent runtime.

▽ Maintenance-mode. This adapter is functional and security-patched, but is not receiving proactive feature work — it is dominated by GenericOpenAICompatModel + a recipe in docs/guides/openai-compat-recipes.md. See maintenance tiers for the rationale. If you actively use this package and want it promoted to ◆ Narrative, open an issue tagged tier:promote-request with your use case.

Install

npm install /model-deepseek /core

Usage

import { DeepSeekModel, DeepSeekModels } from "/model-deepseek";

const model = new DeepSeekModel(DeepSeekModels.V4, {
  apiKey: process.env.DEEPSEEK_API_KEY,
  thinking: { type: "enabled", effort: "high" },
});

⚠️ Compliance — Review the DeepSeek terms of service for data handling and regional access requirements.

License

Apache-2.0 — © wasmagent contributors