DeepSeek Models via OpenRouter

Ever Works does not ship a standalone DeepSeek plugin. DeepSeek models are accessed through the OpenRouter plugin, which is the platform's default AI provider. This page explains how to configure OpenRouter to use DeepSeek models for work generation and AI conversations.

Related source files:

File	Purpose
packages/plugins/openrouter/src/openrouter.plugin.ts	OpenRouter AI provider plugin
packages/plugin/src/ai/reasoning.utils.ts	Reasoning configuration for DeepSeek models
apps/internal-cli/src/commands/config/ai-providers/ai-provider-registry.service.ts	AI provider registry with model lists

Available DeepSeek Models

DeepSeek models are available through two providers in Ever Works:

Via OpenRouter

OpenRouter lists DeepSeek models with the deepseek/ prefix:

Model ID	Description	Best For
deepseek/deepseek-r1	DeepSeek R1 reasoning model	Complex analysis, multi-step reasoning
deepseek/deepseek-chat	DeepSeek Chat	General-purpose content generation
deepseek/deepseek-coder	DeepSeek Coder	Technical content, code-related works

Via Ollama (Local)

DeepSeek models can also be run locally through the Ollama plugin:

Model ID	Description
deepseek-r1	Local DeepSeek R1 (various quantizations)
deepseek-coder-v2	Local DeepSeek Coder V2

Via Groq

Groq provides distilled DeepSeek models with ultra-fast inference:

Model ID	Description
deepseek-r1-distill-qwen-32b	DeepSeek R1 distilled to Qwen 32B

Reasoning Model Configuration

DeepSeek R1 is classified as a reasoning model. The platform automatically applies reasoning configuration to prevent verbose chain-of-thought output from consuming tokens unnecessarily.

From packages/plugin/src/ai/reasoning.utils.ts:

{
    pattern: /deepseek-r|deepseek-reasoner/,
    openrouter: { reasoning: { effort: 'low' } }
}

When a DeepSeek reasoning model is selected, the platform sends a reasoning.effort: 'low' parameter through OpenRouter. This instructs the model to minimize its internal reasoning tokens while still producing structured output.

How Reasoning Config Resolution Works

The extractModelName function strips the provider prefix (deepseek/deepseek-r1 becomes deepseek-r1) before matching against the pattern list.

Configuring DeepSeek as Your AI Provider

Option 1: Through OpenRouter (Recommended)

1. Open the Ever Works dashboard and navigate to Settings > Plugins.
2. Ensure the OpenRouter plugin is enabled (it is by default).
3. Enter your OpenRouter API key if not already configured.
4. Set the model fields to DeepSeek models:

Setting	Recommended Value
Default Model	deepseek/deepseek-chat
Simple Tasks Model	deepseek/deepseek-chat
Standard Tasks Model	deepseek/deepseek-chat
Complex Tasks Model	deepseek/deepseek-r1

Option 2: Through Ollama (Local, Free)

1. Install Ollama on your machine.
2. Pull a DeepSeek model: ollama pull deepseek-r1
3. Enable the Ollama plugin in the Ever Works dashboard.
4. Set the default model to deepseek-r1.

Option 3: Through Groq (Fast Inference)

1. Obtain a Groq API key from console.groq.com.
2. Enable the Groq plugin.
3. Set the model to deepseek-r1-distill-qwen-32b.

Environment Variables

When using DeepSeek through OpenRouter, configure via environment variables:

PLUGIN_OPENROUTER_API_KEY=sk-or-...
PLUGIN_OPENROUTER_DEFAULT_MODEL=deepseek/deepseek-chat
PLUGIN_OPENROUTER_SIMPLE_MODEL=deepseek/deepseek-chat
PLUGIN_OPENROUTER_COMPLEX_MODEL=deepseek/deepseek-r1

Tiered Model Strategy

DeepSeek models work well in a tiered configuration that balances cost and quality:

Tier	Recommended Model	Reasoning
Simple	deepseek/deepseek-chat	Fast responses for tags and short descriptions
Standard	deepseek/deepseek-chat	Solid quality for summaries and listings
Complex	deepseek/deepseek-r1	Chain-of-thought reasoning for detailed pages

Mixing DeepSeek with Other Providers

Because OpenRouter aggregates multiple providers, you can mix DeepSeek models with models from other providers across tiers:

Tier	Model	Provider
Simple	openai/gpt-5-nano	OpenAI
Standard	deepseek/deepseek-chat	DeepSeek
Complex	deepseek/deepseek-r1	DeepSeek

This is configured through the OpenRouter plugin settings -- no additional plugins are needed.

Capabilities

DeepSeek models accessed through OpenRouter support:

Capability	DeepSeek Chat	DeepSeek R1
Structured output	Yes	Yes
Streaming	Yes	Yes
Tool calling	Yes	Limited
Vision	No	No
Embeddings	No	No

note

DeepSeek models do not currently support embeddings. If your workflow requires semantic search, pair DeepSeek with a provider that supports embeddings (OpenAI, Google Gemini, or Ollama).

Troubleshooting

Issue	Cause	Solution
Verbose reasoning output	Reasoning effort not applied	Verify the model ID matches deepseek-r1 or deepseek-reasoner pattern
Model not found	Incorrect model ID	Use the deepseek/ prefix when using OpenRouter
Slow responses with R1	Reasoning models are slower	Use deepseek-chat for simple tasks; reserve R1 for complex ones
Rate limits	OpenRouter rate limiting	Check your OpenRouter plan; add billing credits if needed
Embedding error	DeepSeek does not support embeddings	Switch to OpenAI or Ollama for embedding tasks

Cost Considerations

DeepSeek models are generally more cost-effective than comparable models from OpenAI or Anthropic. When accessed through OpenRouter:

• DeepSeek Chat: Significantly cheaper per token than GPT-4o
• DeepSeek R1: Competitive with other reasoning models while often cheaper

Check openrouter.ai/models for current pricing. OpenRouter displays per-token costs for every available model.

Further Reading

• OpenRouter Plugin -- full OpenRouter configuration reference
• Groq Plugin -- fast inference with distilled DeepSeek models
• Ollama Plugin -- running DeepSeek models locally
• AI Provider Plugins -- overview of all AI provider plugins