DeepSeekAdapter

Built-in adapter for the DeepSeek API. Supports prefix caching, tool use, and streaming.

Import

import { DeepSeekAdapter } from "@halo-ai/adapters";

Constructor

new DeepSeekAdapter(opts: {
  apiKey: string;
  model?: string;       // default: "deepseek-v4-flash"
  baseUrl?: string;     // default: "https://api.deepseek.com"
})

Properties

Property	Value	Description
modelId	"deepseek-v4-flash"	Default model
contextWindow	128_000	Token limit
capabilities.toolUse	true	Tool calling supported
capabilities.streaming	true	Streaming supported
pricing.inputPricePer1k	0.00027	$/1K input tokens
pricing.cachedInputPricePer1k	0.00007	$/1K cached input tokens

Caching

DeepSeek automatically caches the prefix (position 0 in the message array). The adapter sends messages as [...prefix, ...history], placing the stable prefix first for automatic cache detection.