just-deepseek

Rust client for the DeepSeek API.

Quick start

# Cargo.toml
[dependencies]
just-deepseek = "0.1"
tokio = { version = "1", features = ["macros", "rt-multi-thread"] }

use just_deepseek::{
    DeepSeekClient,
    types::chat::{ChatMessage, CreateChatCompletionRequest},
};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let client = DeepSeekClient::new("your-api-key")?;

    let request = CreateChatCompletionRequest::new(
        "deepseek-chat",
        vec![
            ChatMessage::system("You are a concise assistant."),
            ChatMessage::user("Say hello in one sentence."),
        ],
    );

    let response = client.create_chat_completion(request).await?;
    let choice = response.choices.first().expect("expected at least one choice");

    println!("{}", choice.message.content.as_deref().unwrap_or_default());
    Ok(())
}

Highlights

• Streaming — stream_chat_completion() returns a Stream of typed chunks.
• Thinking mode — Enable extended reasoning with ThinkingConfig and ReasoningEffort.
• Tool calling — Full tool-calling loop with ToolDefinition and ToolCallsMessage.
• Wire-level DTOs — Request/response types mirror the upstream DeepSeek API shape.

Looking for a provider-neutral interface? Check out just-llm-client for a unified LLM client with capability traits, runtime provider selection, and tool dispatch.

Ecosystem