chat-deepseek

DeepSeek provider for chat-rs. Thin wrapper over chat-completions targeting DeepSeek's OpenAI-compatible /chat/completions endpoint.

Install

[dependencies]
chat-core = "0.4.0"
chat-deepseek = "0.1.3"
tokio = { version = "1", features = ["macros", "rt-multi-thread"] }

Or via the umbrella crate: chat-rs = { version = "0.5.0", features = ["deepseek"] }.

Usage

use chat_deepseek::DeepSeekBuilder;
use chat_core::{builder::ChatBuilder, types::messages};

let client = DeepSeekBuilder::new()
    .with_model("deepseek-v4-pro")
    .build();

let mut chat = ChatBuilder::new().with_model(client).build();

let mut msgs = messages::from_user(vec!["Hello!"]);
let response = chat.complete(&mut msgs).await?;

Set DEEPSEEK_API_KEY in your environment or call .with_api_key() on the builder. Models: deepseek-v4-pro, deepseek-v4-flash (legacy deepseek-chat / deepseek-reasoner are deprecated).

Capabilities

• Completions — text generation with tool calling and structured output
• Streaming — token-by-token output (requires stream feature)

Note on DSML / Tool-Call Format

DeepSeek V4 open-weight checkpoints natively emit DSML-tagged (XML-ish) tool calls (<｜DSML｜tool_calls> / <｜DSML｜invoke> / <｜DSML｜parameter>). The hosted API parses that into standard OpenAI-shape JSON tool_calls before returning, so this client sees only JSON on the wire. A local DeepSeek runtime (e.g. via chat-mistralrs) would need its own DSML parser at the inference layer — out of scope for this crate.

Custom Endpoint / Transport

Override the base URL with .with_base_url(...) or supply a custom transport with .with_transport(...).

Feature Flags

Streaming is gated on the stream feature:

chat-deepseek = { version = "0.1.2", features = ["stream"] }