ContextQMD
Back to Eliza

DeepSeek Plugin

packages/docs/plugin-registry/llm/deepseek.md

2.0.13.2 KB
Original Source
<Warning> This plugin is not yet available in the Eliza plugin registry. To use DeepSeek models today, configure them through the [OpenRouter plugin](/plugin-registry/llm/openrouter) using the appropriate model ID. </Warning>

The DeepSeek plugin connects Eliza agents to DeepSeek's API, providing access to DeepSeek-V3 (general-purpose) and DeepSeek-R1 (reasoning-focused) models at competitive pricing.

On-demand plugin. This plugin is resolved from the remote elizaOS plugin registry and auto-installs when its API key is detected. It is not included in Eliza's bundled plugins.json index.

Package: @elizaos/plugin-deepseek

Installation

bash
eliza plugins install @elizaos/plugin-deepseek

Auto-Enable

The plugin auto-enables when DEEPSEEK_API_KEY is present:

bash
export DEEPSEEK_API_KEY=sk-...

Configuration

Environment VariableRequiredDescription
DEEPSEEK_API_KEYYesDeepSeek API key from platform.deepseek.com
DEEPSEEK_API_URLNoCustom base URL (default: https://api.deepseek.com)

eliza.json Example

json
{
  "auth": {
    "profiles": {
      "default": {
        "provider": "deepseek",
        "model": "deepseek-chat"
      }
    }
  }
}

Supported Models

ModelContextBest For
deepseek-chat64kGeneral-purpose chat (DeepSeek-V3)
deepseek-reasoner64kChain-of-thought reasoning (DeepSeek-R1)

DeepSeek-V3 is a mixture-of-experts model with 671B parameters (37B active). DeepSeek-R1 is a reasoning model trained with reinforcement learning.

Model Type Mapping

elizaOS Model TypeDeepSeek Model
TEXT_SMALLdeepseek-chat
TEXT_LARGEdeepseek-chat or deepseek-reasoner (configure the large slot)

Features

  • OpenAI-compatible API format
  • Streaming responses
  • Function calling / tool use
  • Multi-turn conversation
  • Code generation (DeepSeek-Coder heritage in V3)
  • Chain-of-thought reasoning (R1)
  • Competitive pricing — significantly cheaper than comparable Western models

DeepSeek-R1 Reasoning

The deepseek-reasoner model produces a <think> block containing its reasoning chain before the final answer. Configure the large text slot to deepseek-reasoner, then use TEXT_LARGE:

typescript
const response = await runtime.useModel("TEXT_LARGE", {
  prompt: "Prove that there are infinitely many prime numbers.",
});

Local DeepSeek via Ollama

For local inference, configure the Ollama plugin with Eliza-1 instead of routing DeepSeek through a local model server.

Rate Limits and Pricing

DeepSeek offers competitive per-token pricing. See platform.deepseek.com/docs/pricing for current rates.

DeepSeek-V3 costs a fraction of GPT-4o at comparable quality for most tasks.