Pi Agent: DeepSeek Configuration

Last updated: 2026-08-31

DeepSeek is Pi Agent's most recommended provider — cheap, capable, excellent with Chinese. This lesson covers the complete DeepSeek configuration.

1. Why DeepSeek

2. Getting an API Key

1. Visit https://platform.deepseek.com/
2. Register and log in
3. Go to "API Keys" page
4. Click "Create API Key"
5. Copy the key (format: sk-xxxxxxxxxxxxxxxx)

Note: The key is shown only once. Save it securely. If compromised, delete it immediately and create a new one.

3. Basic Configuration

(1) Minimal Configuration

providers:
  deepseek:
    api_key: "sk-xxxxxxxx"

(2) Full Configuration

providers:
  deepseek:
    api_key: "sk-xxxxxxxx"
    base_url: "https://api.deepseek.com"
    model: "deepseek-chat"
    temperature: 0.7
    max_tokens: 4096
    top_p: 0.95
    timeout: 60

default_provider: deepseek

(3) Environment Variables

export PI_DEEPSEEK_API_KEY="sk-xxxxxxxx"
export PI_DEEPSEEK_BASE_URL="https://api.deepseek.com"

4. Model Selection

Example 1: Comparing Models (Difficulty: ⭐)

from pi_agent import Agent

chat_agent = Agent(name="quick", provider="deepseek", model="deepseek-chat")
reasoner_agent = Agent(name="thinker", provider="deepseek", model="deepseek-reasoner")

print(chat_agent.chat("1+1=?"))
print(reasoner_agent.chat("Prove that the square root of 2 is irrational"))

5. Parameter Tuning

(1) temperature

Controls output randomness; lower values are more deterministic:

code_agent = Agent(model="deepseek-chat", temperature=0.1)     # Code: precise
creative_agent = Agent(model="deepseek-chat", temperature=1.0) # Creative: diverse

(2) max_tokens

Controls maximum response length:

brief_agent = Agent(model="deepseek-chat", max_tokens=256)     # Short replies
essay_agent = Agent(model="deepseek-chat", max_tokens=8192)    # Long essays

(3) Tool Calling Configuration

agent = Agent(
    provider="deepseek",
    model="deepseek-chat",
    tools=["search", "calculator"],
    tool_choice="auto"  # auto | required | none
)

6. Cost Optimization

Alice processes thousands of requests daily. She uses these strategies:

agent = Agent(
    provider="deepseek",
    model="deepseek-chat",
    max_tokens=2048,
    temperature=0.3,
    context_window=4096
)

Bob advises: "Use deepseek-chat for simple questions, deepseek-reasoner only for complex reasoning. Chat handles 90% of scenarios."

FAQ

Summary

• DeepSeek is the recommended LLM provider for Pi Agent with excellent cost-effectiveness
• Get API key and configure via config file or environment variables
• deepseek-chat for daily use, deepseek-reasoner for complex reasoning
• Optimize with temperature, max_tokens, and tool_choice parameters

Exercises

1. Basic (Difficulty: ⭐): Register a DeepSeek account, get an API key, configure Pi Agent, and send a message.
2. Intermediate (Difficulty: ⭐⭐): Compare deepseek-chat and deepseek-reasoner on the same math problem.
3. Advanced (Difficulty: ⭐⭐⭐): Build a cost monitoring tool that logs token usage and expenses per API call.

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