DeepSeek

DeepSeek's AI models via OpenAI-compatible API.

Configuration

PROVIDER=deepseek
API_KEY=sk-your-deepseek-api-key
MODEL=deepseek-chat
TEMPERATURE=0.3
MAX_TOKENS=1024

Thinking Mode

DeepSeek thinking mode can be controlled with the existing EXTRA_BODY and REASONING_EFFORT settings:

PROVIDER=deepseek
API_KEY=sk-your-deepseek-api-key
MODEL=deepseek-v4-pro
REASONING_EFFORT=high
EXTRA_BODY={"thinking":{"type":"enabled"}}

To explicitly disable thinking mode:

EXTRA_BODY={"thinking":{"type":"disabled"}}

Key Parameters

Parameter	Description	Default
API_KEY	DeepSeek API key (required)	-
MODEL	Model to use	-
TEMPERATURE	Randomness (0.0-1.0)	0.3
TOP_P	Nucleus sampling	1.0
MAX_TOKENS	Max response tokens	1024
TIMEOUT	Request timeout (seconds)	60
BASE_URL	Custom API endpoint	https://api.deepseek.com/v1
EXTRA_BODY	Additional request parameters	{}
REASONING_EFFORT	Thinking effort (high or max for DeepSeek thinking models)	-

Features

• ✅ Streaming responses
• ✅ Function calling
• ✅ MCP support
• ✅ OpenAI-compatible API
• ✅ Code generation
• ✅ Mathematical reasoning

Important Notes

• Uses OpenAI-compatible client implementation
• Supports standard OpenAI API features
• Thinking mode returns reasoning_content, which YAICLI displays as reasoning output
• In thinking mode, DeepSeek ignores temperature, top_p, presence_penalty, and frequency_penalty
• Live tool-call conversations preserve DeepSeek reasoning_content for follow-up requests; saved chat files do not yet persist full reasoning/tool metadata for exact replay after reload
• Designed for coding and reasoning tasks
• Cost-effective alternative to other providers