deepseek

Documentation

Overview

Package deepseek wraps DeepSeek's OpenAI-compatible API.

DeepSeek derives from the OpenAI Chat Completions protocol while retaining its provider-specific reasoning semantics behind Chat.

Provider-specific behavior handled transparently:

• reasoning_content is decoded into a [chat.PartReasoning];
• ordinary prior assistant reasoning is omitted on later turns;
• reasoning associated with tool calls is replayed as reasoning_content.

Provider-specific request controls use the typed RequestOptions extension; the OpenAI SDK request shape is intentionally not exposed. Prefix completion remains a separate beta protocol and is not accepted by Chat.

See https://api-docs.deepseek.com/ for the full API reference.

Index

• Constants
• type Chat
  • func NewChat(config ChatConfig) (*Chat, error)
• type ChatConfig
  • func (c ChatConfig) Validate() error
• type RequestOptions
  • func (r RequestOptions) ValidateFor(generation corechat.Options, tools []corechat.ToolDefinition, stream bool) error
• type ThinkingConfig
  • func (t ThinkingConfig) Validate() error
• type ThinkingMode

Constants

const (
	OpenAIResponseExtensionKey    = "deepseek/openai_response"
	OpenAIStreamChunkExtensionKey = "deepseek/openai_stream_chunk"
)

Namespacing preserves provider-specific data without promoting it into the shared Core protocol or colliding with another provider.

const (
	ModelV4Flash = "deepseek-v4-flash"
	ModelV4Pro   = "deepseek-v4-pro"
)

Production model IDs. See https://api-docs.deepseek.com/quick_start/pricing for the current pricing and context-window limits.

const (
	// BaseURL is DeepSeek's production endpoint. Equivalent to
	// "https://api.deepseek.com/v1" — DeepSeek accepts both.
	BaseURL = "https://api.deepseek.com"
)

const (
	Provider = "DeepSeek"
)

Provider is the stable backend name for host-side attribution.

const RequestExtensionKey = "deepseek/request"

Namespacing preserves provider-specific data without promoting it into the shared Core protocol or colliding with another provider.

Types

type Chat

type Chat = openai.ChatCompletions

Chat implements DeepSeek's chat protocol.

func NewChat

func NewChat(config ChatConfig) (*Chat, error)

NewChat rejects an invalid provider binding before the first chat call.

type ChatConfig

type ChatConfig struct {
	APIKey         string
	DefaultOptions corechat.Options
	BaseURL        string
	HTTPClient     *http.Client
}

ChatConfig binds provider access and defaults shared by every chat call.

func (ChatConfig) Validate

func (c ChatConfig) Validate() error

type RequestOptions

type RequestOptions struct {
	Thinking     *ThinkingConfig `json:"thinking,omitempty"`
	LogProbs     *bool           `json:"logprobs,omitempty"`
	TopLogProbs  *int64          `json:"top_logprobs,omitempty"`
	IncludeUsage *bool           `json:"include_usage,omitempty"`
	UserID       string          `json:"user_id,omitempty"`
}

RequestOptions contains documented DeepSeek Chat Completions fields that have no provider-neutral Core equivalent. Store it in [chat.Options.Extensions] under RequestExtensionKey.

func (RequestOptions) ValidateFor

func (r RequestOptions) ValidateFor(generation corechat.Options, tools []corechat.ToolDefinition, stream bool) error

type ThinkingConfig

type ThinkingConfig struct {
	Type ThinkingMode `json:"type"`
}

ThinkingConfig controls DeepSeek's hybrid thinking mode.

func (ThinkingConfig) Validate

func (t ThinkingConfig) Validate() error

type ThinkingMode

type ThinkingMode string

ThinkingMode selects whether DeepSeek emits reasoning_content before its final answer. The API defaults to enabled when the field is omitted.

const (
	ThinkingEnabled  ThinkingMode = "enabled"
	ThinkingDisabled ThinkingMode = "disabled"
)

These are the provider values this adapter recognizes.