DeepSeek Harness: Model Configuration

Last updated: 2026-08-31

Models are the Agent's "brain" — choose the right model and configure the connection properly, and the Agent can work efficiently. DSH's model-agnostic design lets you freely switch LLMs — from DeepSeek to GPT-4o to local Ollama — by simply modifying configuration with no code changes.

💡 Tip: DSH supports two model access methods: the official DeepSeek API and OpenAI-compatible endpoints. The latter means any service compatible with the OpenAI API format can be connected — including Ollama, vLLM, Azure OpenAI, and more.

📋 Prerequisites: Completed 03-first-use.md, Web UI running normally

1. What You'll Learn

DeepSeek API Key acquisition and configuration
Web UI settings page operations
OpenAI-compatible endpoint configuration
Model routing and multi-model switching
API Key security management best practices

2. DeepSeek API Key Configuration

(1) Obtaining an API Key

Visit the DeepSeek open platform to get an API Key:

Obtain API Key

TEXT 📖 Display only

2. Register/log in to your account
3. Go to the API Keys page
4. Click "Create API Key"
5. Copy the generated Key (starts with sk-)

Note: The API Key is only shown once at creation time. Save it immediately. If lost, you'll need to create a new one.

(2) Configuring the API Key

Three configuration methods, in order of priority from high to low:

Method 1: Web UI Settings Page (Recommended)

TEXT 📖 Display only

Top control bar → ⚙️ Settings → Models → DeepSeek API Key
┌─────────────────────────────────────────┐
│ DeepSeek API Key                         │
│ sk-•••••••••••••••••••••••••••xxxx       │
│ [Test Connection]  [Save]                │
└─────────────────────────────────────────┘

Click Test Connection to verify connectivity:

Configure API Key

TEXT 📖 Display only

✅ Connection successful!
   Model: deepseek-chat
   Latency: 120ms

Method 2: Environment Variable

BASH

# Configure in .env file or shell
export DEEPSEEK_API_KEY="sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"

Method 3: Configuration File

YAML

# dsh.config.yaml
models:
  deepseek:
    api_key: "sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
    base_url: "https://api.deepseek.com/v1"

(3) Available DeepSeek Models

Model ID	Description	Context Window	Use Cases
`deepseek-chat`	General conversation model	64K tokens	Daily programming, Q&A
`deepseek-reasoner`	Reasoning-enhanced model	64K tokens	Complex logic, mathematical reasoning
`deepseek-coder`	Code-specific model	16K tokens	Code generation, debugging

3. Settings Page Details

(1) ▶ Example 1

Settings Page

TEXT 📖 Display only

┌──────────────────────────────────────────────┐
│ 📡 Model Providers                            │
│                                                │
│ ┌─ DeepSeek ────────────────────────────────┐ │
│ │ API Key: sk-•••••••••xxxx                 │ │
│ │ Base URL: https://api.deepseek.com/v1     │ │
│ │ Default Model: deepseek-chat              │ │
│ │ [Test] [Save]                             │ │
│ └───────────────────────────────────────────┘ │
│                                                │
│ ┌─ OpenAI Compatible ───────────────────────┐ │
│ │ API Key: sk-•••••••••xxxx                 │ │
│ │ Base URL: https://api.openai.com/v1       │ │
│ │ Default Model: gpt-4o                     │ │
│ │ [Test] [Save]                             │ │
│ └───────────────────────────────────────────┘ │
│                                                │
│ [+ Add Provider]                               │
└──────────────────────────────────────────────┘

(2) Configuration Item Descriptions

Configuration Item	Description	Example
API Key	LLM service authentication key	`sk-xxxxxxxx`
Base URL	API endpoint address	`https://api.deepseek.com/v1`
Default Model	Default model to use	`deepseek-chat`
Max Tokens	Maximum tokens per response	`4096`
Temperature	Generation temperature (0-2)	`0.7`

(3) ▶ Example 3

YAML

# dsh.config.yaml
models:
  deepseek:
    api_key: "${DEEPSEEK_API_KEY}"
    default_model: deepseek-chat
    params:
      temperature: 0.7
      max_tokens: 4096
      top_p: 0.95
  openai:
    api_key: "${OPENAI_API_KEY}"
    default_model: gpt-4o
    params:
      temperature: 0.5
      max_tokens: 8192

Recommended parameters for different scenarios:

Scenario	Temperature	Max Tokens	Notes
Code generation	0.2-0.4	4096+	Low temperature ensures code consistency
Documentation writing	0.5-0.7	4096	Moderate creativity
Creative brainstorming	0.8-1.0	8192+	High temperature encourages diversity
Precise Q&A	0.0-0.2	2048	Most deterministic output

4. OpenAI-Compatible Endpoint Configuration

(1) ▶ Example 1

YAML

# dsh.config.yaml
models:
  openai:
    api_key: "${OPENAI_API_KEY}"
    base_url: "https://api.openai.com/v1"
    default_model: gpt-4o

(2) Azure OpenAI Endpoint

YAML

# dsh.config.yaml
models:
  azure:
    api_key: "${AZURE_OPENAI_API_KEY}"
    base_url: "https://your-resource.openai.azure.com/openai/deployments/your-deployment"
    default_model: gpt-4o
    headers:
      api-key: "${AZURE_OPENAI_API_KEY}"

(3) Local Ollama Endpoint

YAML

# dsh.config.yaml
models:
  ollama:
    api_key: "ollama"           # Ollama doesn't need a real Key
    base_url: "http://localhost:11434/v1"
    default_model: llama3.2

(4) Custom Endpoint

Any service compatible with the OpenAI API format can be connected:

YAML

# dsh.config.yaml
models:
  custom:
    api_key: "${CUSTOM_API_KEY}"
    base_url: "https://your-llm-service.com/v1"
    default_model: your-model-name
    headers:
      X-Custom-Header: "custom-value"

Plugin Architecture

LLM Adapter

(5) Multi-Endpoint Architecture

graph TB
    DSH[DeepSeek Harness] -->|Provider 1| DS[DeepSeek API<br/>deepseek-chat]
    DSH -->|Provider 2| OAI[OpenAI<br/>gpt-4o]
    DSH -->|Provider 3| OLL[Ollama<br/>llama3.2]
    DSH -->|Provider 4| AZ[Azure OpenAI<br/>gpt-4o]

5. Model Routing and Multi-Model Selection

(1) Switching Models in Web UI

Switch models using the dropdown in the top control bar:

TEXT 📖 Display only

[deepseek-chat ▼]
┌──────────────────────────┐
│ ● deepseek-chat          │
│   DeepSeek · 64K ctx     │
│                          │
│ ○ gpt-4o                 │
│   OpenAI · 128K ctx      │
│                          │
│ ○ llama3.2               │
│   Ollama · 8K ctx        │
└──────────────────────────┘

(2) Model Routing Strategy

DSH supports automatically routing to different models based on task type:

YAML

# dsh.config.yaml
routing:
  rules:
    - match:
        mode: ptc            # PTC mode
      model: deepseek-reasoner   # Use reasoning model for planning
    
    - match:
        tool: file_edit      # File editing
      model: deepseek-coder  # Use code model
    
    - match:
        mode: creative       # Creative mode
      model: gpt-4o          # Use GPT-4o
    
    - match:
        default: true        # Default route
      model: deepseek-chat

(3) Routing Flow

graph TD
    INPUT[User Input] --> CHECK{Match Routing Rules}
    CHECK -->|PTC Mode| R1[deepseek-reasoner]
    CHECK -->|file_edit Tool| R2[deepseek-coder]
    CHECK -->|Creative Mode| R3[gpt-4o]
    CHECK -->|Default| R4[deepseek-chat]

(4) Specifying Models in CLI Mode

BASH

# Specify model at startup
npx @deepseek-ai/dsh cli --model gpt-4o

# Switch models during a session (CLI internal command)
/model deepseek-reasoner

6. API Key Security Management

(1) Security Principles

Principle	Description
No hardcoding	API Keys should not be written directly in code
Don't commit to Git	Add `.env` file to `.gitignore`
Least privilege	Only configure necessary service access permissions
Regular rotation	Periodically replace API Keys
Audit logs	Monitor API Key usage

(2) Environment Variable Management

BASH

# .env file (not committed to Git)
DEEPSEEK_API_KEY=sk-xxxxxxxxxxxxxxxx
OPENAI_API_KEY=sk-yyyyyyyyyyyyyyyy

# .env.example file (committed to Git, for team reference)
DEEPSEEK_API_KEY=your-deepseek-api-key
OPENAI_API_KEY=your-openai-api-key

BASH

# .gitignore
.env
.env.local
.env.*.local

(3) Variable References in Configuration Files

DSH configuration files support environment variable references to avoid hardcoding:

YAML

# dsh.config.yaml — Use ${VAR} to reference environment variables
models:
  deepseek:
    api_key: "${DEEPSEEK_API_KEY}"    # Read from environment variable at runtime
    base_url: "https://api.deepseek.com/v1"
  openai:
    api_key: "${OPENAI_API_KEY}"
    base_url: "https://api.openai.com/v1"

(4) Secret Management Service Integration

For enterprise deployments, professional secret management services are recommended:

Key Management

YAML

models:
  deepseek:
    api_key:
      vault: aws-secrets-manager
      secret_id: "dsh/deepseek-api-key"
      region: "us-east-1"

(5) Key Leak Emergency Response

TEXT 📖 Display only

1. Immediately disable/delete the leaked Key on the API provider's platform
2. Generate a new Key and update configuration
3. Check API call logs to confirm no abnormal usage
4. Review Git history to ensure the Key hasn't been committed (if committed, use git filter-branch to remove it)

7. Connection Troubleshooting

(1) Common Errors and Solutions

Error Message	Cause	Solution
`Invalid API key`	Key is wrong or expired	Regenerate Key and configure

Connection Troubleshooting

(2) Connection Test Commands

BASH

# Test DeepSeek API
curl -s https://api.deepseek.com/v1/models \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY"

# Test OpenAI-compatible endpoint
curl -s https://api.openai.com/v1/models \
  -H "Authorization: Bearer $OPENAI_API_KEY"

# Test local Ollama
curl -s http://localhost:11434/v1/models

(3) Proxy Configuration

If you need to access the API through a proxy:

BASH

# Set HTTP proxy
export HTTP_PROXY="http://proxy.example.com:8080"
export HTTPS_PROXY="http://proxy.example.com:8080"

# Or specify in configuration file
# dsh.config.yaml
models:
  deepseek:
    api_key: "${DEEPSEEK_API_KEY}"
    base_url: "https://api.deepseek.com/v1"
    proxy: "http://proxy.example.com:8080"

❓ FAQ

Q Can I configure multiple model providers simultaneously?

A Yes. DSH supports configuring DeepSeek, OpenAI, Ollama, and other providers simultaneously, with one-click switching in the Web UI.

Q Is the free API Key enough?

A DeepSeek new users typically get free credits, suitable for learning and testing. However, for long-term use, we recommend purchasing a formal plan; free credits may not be sufficient for complex Agent tasks.

Q How do I know which model is currently in use?

A The Web UI top control bar displays the current model name. In CLI mode, enter /model to view the current model.

Q Is context preserved after switching models?

A Session context (conversation history) is preserved, but different models have different context window sizes — if history exceeds the window, earlier messages will be truncated.

Q Are Ollama local models effective?

A It depends on model size and quantization level. 7B parameter models are suitable for simple tasks; 70B+ models approach cloud performance but require significant GPU resources.

Q My API Key is configured but I still get connection errors. What should I do?

A Troubleshoot in order: 1) Is the Key correct (no extra spaces); 2) Is the Base URL correct; 3) Is the network connected; 4) Is a proxy needed.

📖 Summary

DeepSeek API Key can be configured via environment variables, configuration files, or Web UI settings
OpenAI-compatible endpoints let DSH connect to any OpenAI API format service
Model routing strategies can automatically select the optimal model based on mode/tool
API Key security management: no hardcoding, don't commit to Git, use environment variable references
Switch models with one click at the Web UI top; use /model command in CLI
Troubleshoot connection issues in Key → URL → Network → Proxy order

📝 Exercises

1. ⭐ Basic： Configure a DeepSeek API Key, successfully send a message in the Web UI and receive a reply. Save a screenshot of the connection test results.

2. ⭐⭐ Intermediate： Configure both DeepSeek and OpenAI model providers, switch between the two models in the same session, and compare the differences in responses to the same question.

3. ⭐⭐⭐ Challenge： Configure an Ollama local model endpoint, and write model routing rules — route code editing operations to DeepSeek Coder, creative tasks to GPT-4o, and use DeepSeek Chat as the default. Verify that routing works as expected.

Web-Tutorial Tech Team

A team of developers maintaining programming tutorials. Each tutorial is written and reviewed by developers with expertise in that field. We work to keep our content accurate and reliable — if you spot an issue, please let us know.

About Us

DeepSeek Harness: Model Configuration

1. What You'll Learn

2. DeepSeek API Key Configuration

(1) Obtaining an API Key

(2) Configuring the API Key

(3) Available DeepSeek Models

3. Settings Page Details

(1) ▶ Example 1

(2) Configuration Item Descriptions

(3) ▶ Example 3

4. OpenAI-Compatible Endpoint Configuration

(1) ▶ Example 1

(2) Azure OpenAI Endpoint

(3) Local Ollama Endpoint

(4) Custom Endpoint

(5) Multi-Endpoint Architecture

5. Model Routing and Multi-Model Selection

(1) Switching Models in Web UI

(2) Model Routing Strategy

(3) Routing Flow

(4) Specifying Models in CLI Mode

6. API Key Security Management

(1) Security Principles

(2) Environment Variable Management

(3) Variable References in Configuration Files

(4) Secret Management Service Integration

(5) Key Leak Emergency Response

7. Connection Troubleshooting

(1) Common Errors and Solutions

(2) Connection Test Commands

(3) Proxy Configuration

❓ FAQ

📖 Summary

📝 Exercises

🙏 帮我们做得更好