just-deepseek

Wire-level types, a thin async client, and HTTP transport helpers for the DeepSeek API.

Overview

This crate provides serde-serializable request and response types that closely mirror the DeepSeek HTTP API shape, a thin async client (DeepSeekClient), and re-exported HTTP transport helpers for making requests with a raw reqwest::Client.

• types — Wire-level DTOs (request/response types with serde derives).
• transport — Re-exported HTTP helpers from just-common (build_client, endpoint_url, ensure_success, parse_json, JsonEventStream).

For a provider-neutral abstraction with capability traits and runtime provider selection, use just-llm-client with the deepseek feature.

Quick start

Type serialization

use just_deepseek::types::chat::{ChatCompletionRequest, ChatMessage};

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

let json = serde_json::to_string(&request)?;

HTTP request with transport helpers

use just_deepseek::transport::{build_client, endpoint_url, ensure_success};
use just_deepseek::types::chat::{ChatCompletion, ChatCompletionRequest, ChatMessage};

let http = build_client(
    reqwest::Client::builder().use_rustls_tls(),
    "your-api-key",
)?;
let request = ChatCompletionRequest::new(
    "deepseek-chat",
    vec![
        ChatMessage::system("You are a concise assistant."),
        ChatMessage::user("Say hello in one sentence."),
    ],
);

let url = endpoint_url("https://api.deepseek.com", "/chat/completions")?;
let response = http
    .post(&url)
    .header(reqwest::header::CONTENT_TYPE, "application/json")
    .body(serde_json::to_vec(&request)?)
    .send()
    .await?;
let response = ensure_success(response).await?;
let completion: ChatCompletion = response.json().await?;

Highlights

• Wire-level DTOs — Request/response types mirror the upstream DeepSeek API shape with serde derives.
• Thinking mode — ThinkingConfig and ReasoningEffort fields for extended reasoning.
• Tool calling — ToolDefinition, ToolCallsMessage, and related types for tool-calling workflows.
• Transport helpers — Re-exported HTTP utilities from just-common for convenience. Use just-llm-client for a full client abstraction.

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

Examples

# Non-streaming chat completion
JUST_LLM_DEEPSEEK_API_KEY=your-key JUST_LLM_DEEPSEEK_MODEL=deepseek-chat \
  cargo run -p just-deepseek --example chat_completion

# Streaming
JUST_LLM_DEEPSEEK_API_KEY=your-key JUST_LLM_DEEPSEEK_MODEL=deepseek-chat \
  cargo run -p just-deepseek --example streaming_chat_completion

# Thinking mode (DeepSeek-specific)
JUST_LLM_DEEPSEEK_API_KEY=your-key JUST_LLM_DEEPSEEK_MODEL=deepseek-chat \
  cargo run -p just-deepseek --example thinking_mode

# Tool calling loop
cargo run -p just-deepseek --example tool_calling

Ecosystem