Documentation
¶
Index ¶
- Constants
- Variables
- func ErrBatchUnsupported() error
- func IsAzure(apiType ProviderType) bool
- func IsBedrock(apiType ProviderType) bool
- type BatchCreateParams
- type BatchEndpoint
- type ChatCompletionChoice
- type ChatCompletionResponse
- type ChatMessage
- type ChatRequest
- type ChatUsage
- type Client
- func (c *Client) CancelBatch(ctx context.Context, batchID string) (*openaisdk.Batch, error)
- func (c *Client) CreateBatch(ctx context.Context, params BatchCreateParams) (*openaisdk.Batch, error)
- func (c *Client) CreateChat(ctx context.Context, r *ChatRequest) (*ChatCompletionResponse, error)
- func (c *Client) CreateCompletion(ctx context.Context, r *CompletionRequest) (*Completion, error)
- func (c *Client) CreateEmbedding(ctx context.Context, r *EmbeddingRequest) ([][]float32, error)
- func (c *Client) CreateResponse(ctx context.Context, r *responses.ResponseNewParams) (*responses.Response, error)
- func (c *Client) CreateStreamingResponse(ctx context.Context, r *responses.ResponseNewParams, ...) (*responses.Response, error)
- func (c *Client) DownloadFile(ctx context.Context, fileID string) (io.ReadCloser, error)
- func (c *Client) GetBatch(ctx context.Context, batchID string) (*openaisdk.Batch, error)
- func (c *Client) InferChat(ctx context.Context, body map[string]any) (*llms.InferenceResponse, error)
- func (c *Client) InferResponses(ctx context.Context, body map[string]any) (*llms.InferenceResponse, error)
- func (c *Client) ListModels(ctx context.Context) ([]llms.ModelInfo, error)
- func (c *Client) SupportsResponsesAPI() bool
- func (c *Client) UploadBatchFile(ctx context.Context, name string, body io.Reader) (string, error)
- type Completion
- type CompletionRequest
- type CompletionResponse
- type Doer
- type EmbeddingRequest
- type FinishReason
- type FunctionCall
- type FunctionCallBehavior
- type FunctionDefinition
- type LogProb
- type LogProbs
- type Option
- type ProviderType
- type StreamOptions
- type StreamedChatResponsePayload
- type Tool
- type ToolCall
- type ToolChoice
- type ToolFunction
- type ToolType
- type TopLogProbs
- type Usage
- type WebSearchOptions
Constants ¶
const ( ResponsesItemMessage = "message" ResponsesItemFunctionCall = "function_call" ResponsesItemFunctionOutput = "function_call_output" ResponsesItemReasoning = "reasoning" ResponsesPartInputText = "input_text" ResponsesPartOutputText = "output_text" ResponsesPartRefusal = "refusal" ResponsesPartSummaryText = "summary_text" )
Responses API wire vocabulary.
const ( DefaultBaseURL = "https://api.openai.com/v1" DefaultFunctionCallBehavior = "auto" DefaultChatModel = "gpt-5-mini" DefaultMaxTokens = 2 * 16384 )
Variables ¶
var ErrContentExclusive = errors.New("only one of Content / MultiContent allowed in message")
var ErrEmptyResponse = errors.New("empty response")
ErrEmptyResponse is returned when the OpenAI API returns an empty response.
Functions ¶
func ErrBatchUnsupported ¶ added in v0.17.130
func ErrBatchUnsupported() error
ErrBatchUnsupported is the package-level sentinel exposed for upstream callers that want to detect "provider does not support batching".
func IsAzure ¶
func IsAzure(apiType ProviderType) bool
func IsBedrock ¶ added in v0.19.148
func IsBedrock(apiType ProviderType) bool
Types ¶
type BatchCreateParams ¶ added in v0.17.130
type BatchCreateParams struct {
// InputFileID is the file_id returned by UploadBatchFile.
InputFileID string
// Endpoint is the OpenAI URL path used by every request in the batch.
Endpoint BatchEndpoint
// CompletionWindow is the maximum time the batch may run. Currently only
// "24h" is accepted by the API; empty defaults to "24h".
CompletionWindow string
// Metadata is round-tripped through the API on the batch object.
Metadata map[string]string
}
BatchCreateParams is the input to CreateBatch.
type BatchEndpoint ¶ added in v0.17.130
type BatchEndpoint string
BatchEndpoint is the OpenAI URL path used by every request in a batch.
const ( // EndpointChatCompletions targets the /v1/chat/completions endpoint. EndpointChatCompletions BatchEndpoint = "/v1/chat/completions" // EndpointResponses targets the /v1/responses endpoint. EndpointResponses BatchEndpoint = "/v1/responses" )
type ChatCompletionChoice ¶
type ChatCompletionChoice struct {
Index int `json:"index"`
Message ChatMessage `json:"message"`
FinishReason FinishReason `json:"finish_reason"`
LogProbs *LogProbs `json:"logprobs,omitempty"`
}
ChatCompletionChoice is a choice in a chat response.
type ChatCompletionResponse ¶
type ChatCompletionResponse struct {
ID string `json:"id,omitempty"`
Created int64 `json:"created,omitempty"`
Choices []*ChatCompletionChoice `json:"choices,omitempty"`
Model string `json:"model,omitempty"`
Object string `json:"object,omitempty"`
Usage ChatUsage `json:"usage,omitempty"`
SystemFingerprint string `json:"system_fingerprint"`
}
ChatCompletionResponse is a response to a chat request.
type ChatMessage ¶
type ChatMessage struct {
// The role of the author of this message. One of system, user, assistant, function, or tool.
Role string
// The content of the message.
// This field is mutually exclusive with MultiContent.
Content string
// MultiContent is a list of content parts to use in the message.
MultiContent []llms.ContentPart
// The name of the author of this message. May contain a-z, A-Z, 0-9, and underscores,
// with a maximum length of 64 characters.
Name string
// ToolCalls is a list of tools that were called in the message.
ToolCalls []ToolCall `json:"tool_calls,omitempty"`
// ToolCallID is the ID of the tool call this message is for.
// Only present in tool messages.
ToolCallID string `json:"tool_call_id,omitempty"`
// This field is only used with the deepseek-reasoner model and represents the reasoning contents of the assistant message before the final answer.
ReasoningContent string `json:"reasoning_content,omitempty"`
}
ChatMessage is a message in a chat request.
func (ChatMessage) MarshalJSON ¶
func (m ChatMessage) MarshalJSON() ([]byte, error)
func (*ChatMessage) UnmarshalJSON ¶
func (m *ChatMessage) UnmarshalJSON(data []byte) error
type ChatRequest ¶
type ChatRequest struct {
Model string `json:"model"`
Messages []*ChatMessage `json:"messages"`
Temperature float64 `json:"temperature"`
TopP float64 `json:"top_p,omitempty"`
MaxCompletionTokens int `json:"max_completion_tokens,omitempty"`
N int `json:"n,omitempty"`
StopWords []string `json:"stop,omitempty"`
Stream bool `json:"stream,omitempty"`
FrequencyPenalty float64 `json:"frequency_penalty,omitempty"`
PresencePenalty float64 `json:"presence_penalty,omitempty"`
Seed int `json:"seed,omitempty"`
// ReasoningEffort is the reasoning effort for models that have such a
// control. Empty leaves the model's own default in place.
ReasoningEffort string `json:"reasoning_effort,omitempty"`
// ResponseFormat is the format of the response.
ResponseFormat *schema.ResponseFormat `json:"response_format,omitempty"`
// LogProbs indicates whether to return log probabilities of the output tokens or not.
// If true, returns the log probabilities of each output token returned in the content of message.
// This option is currently not available on the gpt-4-vision-preview model.
LogProbs bool `json:"logprobs,omitempty"`
// TopLogProbs is an integer between 0 and 5 specifying the number of most likely tokens to return at each
// token position, each with an associated log probability.
// logprobs must be set to true if this parameter is used.
TopLogProbs int `json:"top_logprobs,omitempty"`
Tools []Tool `json:"tools,omitempty"`
// This can be either a string or a ToolChoice object.
// If it is a string, it should be one of 'none', or 'auto', otherwise it should be a ToolChoice object specifying a specific tool to use.
ToolChoice any `json:"tool_choice,omitempty"`
// Options for streaming response. Only set this when you set stream: true.
StreamOptions *StreamOptions `json:"stream_options,omitempty"`
// StreamingFunc is a function to be called for each chunk of a streaming response.
// Return an error to stop streaming early.
StreamingFunc func(ctx context.Context, chunk []byte) error `json:"-"`
// StreamingReasoningFunc is a function to be called for each reasoning and content chunk of a streaming response.
// Return an error to stop streaming early.
StreamingReasoningFunc func(ctx context.Context, reasoningChunk, chunk []byte) error `json:"-"`
// Metadata allows you to specify additional information that will be passed to the model.
Metadata map[string]any `json:"metadata,omitempty"`
// PromptCacheKey identifies a cached prompt on providers that support prompt caching.
PromptCacheKey string `json:"prompt_cache_key,omitempty"`
// PromptCacheRetention controls request-level prompt cache retention.
PromptCacheRetention string `json:"prompt_cache_retention,omitempty"`
}
ChatRequest is a request to complete a chat completion..
type ChatUsage ¶
type ChatUsage struct {
PromptTokens int `json:"prompt_tokens"`
CompletionTokens int `json:"completion_tokens"`
TotalTokens int `json:"total_tokens"`
CompletionTokensDetails struct {
ReasoningTokens int `json:"reasoning_tokens"`
} `json:"completion_tokens_details"`
}
ChatUsage is the usage of a chat completion request.
type Client ¶
type Client struct {
Model string
Provider ProviderType
EmbeddingModel string
ResponseFormat *schema.ResponseFormat
// contains filtered or unexported fields
}
Client is a client for the OpenAI API.
func New ¶
func New(provider ProviderType, model string, token string, baseURL string, organization, project string, apiVersion string, httpClient Doer, embeddingModel string, responseFormat *schema.ResponseFormat, headers map[string]string, opts ...Option, ) (*Client, error)
New returns a new OpenAI client.
func (*Client) CancelBatch ¶ added in v0.17.130
CancelBatch requests cancellation of an in-flight batch.
func (*Client) CreateBatch ¶ added in v0.17.130
func (c *Client) CreateBatch(ctx context.Context, params BatchCreateParams) (*openaisdk.Batch, error)
CreateBatch creates a new batch from a previously uploaded input file.
func (*Client) CreateChat ¶
func (c *Client) CreateChat(ctx context.Context, r *ChatRequest) (*ChatCompletionResponse, error)
CreateChat creates chat request.
func (*Client) CreateCompletion ¶
func (c *Client) CreateCompletion(ctx context.Context, r *CompletionRequest) (*Completion, error)
CreateCompletion creates a completion.
func (*Client) CreateEmbedding ¶
CreateEmbedding creates embeddings.
func (*Client) CreateResponse ¶ added in v0.14.88
func (c *Client) CreateResponse(ctx context.Context, r *responses.ResponseNewParams) (*responses.Response, error)
CreateResponse creates a response using the Responses API.
func (*Client) CreateStreamingResponse ¶ added in v0.15.112
func (c *Client) CreateStreamingResponse( ctx context.Context, r *responses.ResponseNewParams, streamFunc func(ctx context.Context, chunk []byte) error, ) (*responses.Response, error)
CreateStreamingResponse creates a response using the Responses API with SSE streaming. streamFunc is called for each text delta chunk. The full Response (with usage stats) is returned once the response.completed event is received.
func (*Client) DownloadFile ¶ added in v0.17.130
DownloadFile streams the raw contents of a file (typically a batch output or error file). The caller is responsible for closing the returned reader.
func (*Client) GetBatch ¶ added in v0.17.130
GetBatch retrieves the current state of a submitted batch.
func (*Client) InferChat ¶ added in v0.21.160
func (c *Client) InferChat(ctx context.Context, body map[string]any) (*llms.InferenceResponse, error)
InferChat performs one non-streaming Chat Completions call with the raw body supplied by the connector, using the shared authenticated SDK transport and no retries. Legacy defaults are not changed.
func (*Client) InferResponses ¶ added in v0.21.160
func (c *Client) InferResponses(ctx context.Context, body map[string]any) (*llms.InferenceResponse, error)
InferResponses performs one non-streaming Responses API call with the raw body supplied by the connector, using the shared authenticated SDK transport and no retries. Reasoning items are returned with their raw JSON as opaque state.
func (*Client) ListModels ¶ added in v1.0.173
ListModels returns the models the provider lists for the account, in the provider's order. It reads the fields of the OpenAI list, and the extra ones of the OpenRouter list when they are present:
shutdown_date, expiration_date a date makes the model deprecated,
null makes it active
name the display name
reasoning.supported_efforts the reasoning effort levels
reasoning.default_effort the level applied when none is named
A reasoning level without a portable name, such as "minimal", is left out.
Neither API pages its list today. A list that says it holds more models than it returned is an error, because a truncated list would make models look retired.
func (*Client) SupportsResponsesAPI ¶ added in v0.14.88
func (*Client) UploadBatchFile ¶ added in v0.17.130
UploadBatchFile uploads a JSONL file with purpose=batch and returns the file ID that should be passed to CreateBatch. The caller is responsible for formatting body as a JSONL document where each line is a valid batch request.
type CompletionRequest ¶
type CompletionRequest struct {
Model string `json:"model"`
Prompt string `json:"prompt"`
Temperature float64 `json:"temperature"`
MaxCompletionTokens int `json:"max_completion_tokens,omitempty"`
N int `json:"n,omitempty"`
FrequencyPenalty float64 `json:"frequency_penalty,omitempty"`
PresencePenalty float64 `json:"presence_penalty,omitempty"`
TopP float64 `json:"top_p,omitempty"`
StopWords []string `json:"stop,omitempty"`
Seed int `json:"seed,omitempty"`
// StreamingFunc is a function to be called for each chunk of a streaming response.
// Return an error to stop streaming early.
StreamingFunc func(ctx context.Context, chunk []byte) error `json:"-"`
}
CompletionRequest is a request to complete a completion.
type CompletionResponse ¶
type CompletionResponse struct {
ID string `json:"id,omitempty"`
Created float64 `json:"created,omitempty"`
Choices []struct {
FinishReason string `json:"finish_reason,omitempty"`
Index float64 `json:"index,omitempty"`
Logprobs any `json:"logprobs,omitempty"`
Text string `json:"text,omitempty"`
} `json:"choices,omitempty"`
Model string `json:"model,omitempty"`
Object string `json:"object,omitempty"`
Usage struct {
CompletionTokens float64 `json:"completion_tokens,omitempty"`
PromptTokens float64 `json:"prompt_tokens,omitempty"`
TotalTokens float64 `json:"total_tokens,omitempty"`
} `json:"usage,omitempty"`
}
type EmbeddingRequest ¶
EmbeddingRequest is a request to create an embedding.
type FinishReason ¶
type FinishReason string
const ( FinishReasonStop FinishReason = "stop" FinishReasonLength FinishReason = "length" FinishReasonFunctionCall FinishReason = "function_call" FinishReasonToolCalls FinishReason = "tool_calls" FinishReasonContentFilter FinishReason = "content_filter" FinishReasonNull FinishReason = "null" )
func (FinishReason) MarshalJSON ¶
func (r FinishReason) MarshalJSON() ([]byte, error)
type FunctionCall ¶
type FunctionCall struct {
// Name is the name of the function to call.
Name string `json:"name"`
// Arguments is the set of arguments to pass to the function.
Arguments string `json:"arguments"`
}
FunctionCall is a call to a function.
type FunctionCallBehavior ¶
type FunctionCallBehavior string
FunctionCallBehavior is the behavior to use when calling functions.
const ( // FunctionCallBehaviorUnspecified is the empty string. FunctionCallBehaviorUnspecified FunctionCallBehavior = "" // FunctionCallBehaviorNone will not call any functions. FunctionCallBehaviorNone FunctionCallBehavior = "none" // FunctionCallBehaviorAuto will call functions automatically. FunctionCallBehaviorAuto FunctionCallBehavior = "auto" )
type FunctionDefinition ¶
type FunctionDefinition struct {
// Name is the name of the function.
Name string `json:"name"`
// Description is a description of the function.
Description string `json:"description,omitempty"`
// Parameters is the JSON Schema of the function arguments. Nil leaves
// it out of the request: the function takes no arguments.
Parameters any `json:"parameters,omitempty"`
// Strict is a flag to enable structured output mode.
Strict bool `json:"strict,omitempty"`
}
FunctionDefinition is a definition of a function that can be called by the model.
type LogProb ¶
type LogProb struct {
Token string `json:"token"`
LogProb float64 `json:"logprob"`
Bytes []byte `json:"bytes,omitempty"` // Omitting the field if it is null
// TopLogProbs is a list of the most likely tokens and their log probability, at this token position.
// In rare cases, there may be fewer than the number of requested top_logprobs returned.
TopLogProbs []TopLogProbs `json:"top_logprobs"`
}
LogProb represents the probability information for a token.
type LogProbs ¶
type LogProbs struct {
// Content is a list of message content tokens with log probability information.
Content []LogProb `json:"content"`
}
LogProbs is the top-level structure containing the log probability information.
type ProviderType ¶ added in v0.14.88
type ProviderType string
const ( ProviderOpenAI ProviderType = "OPENAI" ProviderOpenAIBedrock ProviderType = "OPENAI_BEDROCK" ProviderOpenRouter ProviderType = "OPENROUTER" ProviderAzure ProviderType = "AZURE" ProviderAzureAD ProviderType = "AZURE_AD" )
type StreamOptions ¶
type StreamOptions struct {
// If set, an additional chunk will be streamed before the data: [DONE] message.
// The usage field on this chunk shows the token usage statistics for the entire request,
// and the choices field will always be an empty array.
// All other chunks will also include a usage field, but with a null value.
IncludeUsage bool `json:"include_usage,omitempty"`
}
type StreamedChatResponsePayload ¶
type StreamedChatResponsePayload struct {
ID string `json:"id,omitempty"`
Created float64 `json:"created,omitempty"`
Model string `json:"model,omitempty"`
Object string `json:"object,omitempty"`
Choices []struct {
Index float64 `json:"index,omitempty"`
Delta struct {
Role string `json:"role,omitempty"`
Content string `json:"content,omitempty"`
// ToolCalls is a list of tools that were called in the message.
ToolCalls []*ToolCall `json:"tool_calls,omitempty"`
// This field is only used with the deepseek-reasoner model and represents the reasoning contents of the assistant message before the final answer.
ReasoningContent string `json:"reasoning_content,omitempty"`
} `json:"delta,omitempty"`
FinishReason FinishReason `json:"finish_reason,omitempty"`
} `json:"choices,omitempty"`
SystemFingerprint string `json:"system_fingerprint"`
// An optional field that will only be present when you set stream_options: {"include_usage": true} in your request.
// When present, it contains a null value except for the last chunk which contains the token usage statistics
// for the entire request.
Usage *Usage `json:"usage,omitempty"`
Error error `json:"-"` // use for error handling only
}
StreamedChatResponsePayload is a chunk from the stream.
type Tool ¶
type Tool struct {
Type ToolType `json:"type"`
Function *FunctionDefinition `json:"function,omitempty"`
// For tools like web_search, you can include extra options.
WebSearchOptions *WebSearchOptions `json:"-"`
}
Tool is a tool to use in a chat request.
func (Tool) MarshalJSON ¶ added in v0.14.88
type ToolCall ¶
type ToolCall struct {
ID string `json:"id,omitempty"`
Type ToolType `json:"type"`
Function ToolFunction `json:"function,omitempty"`
}
ToolCall is a call to a tool.
type ToolChoice ¶
type ToolChoice struct {
Type ToolType `json:"type"`
Function ToolFunction `json:"function,omitempty"`
}
ToolChoice is a choice of a tool to use.
type ToolFunction ¶
ToolFunction is a function to be called in a tool choice.
type TopLogProbs ¶
type WebSearchOptions ¶ added in v0.14.88
type WebSearchOptions struct {
AllowedDomains []string
}