Documentation
¶
Overview ¶
Package openai provides an implementation of the Provider interface for OpenAI.
v0.10.0 #166 — OpenAI ReasoningProvider implementation via the Responses API.
OpenAI's reasoning models (o3, o4-mini, gpt-5 reasoning class) expose reasoning summaries through the Responses API at /v1/responses, NOT through Chat Completions. The Responses API accepts an "input" array (role + content strings) and returns an "output" array containing alternating "reasoning" and "message" items. Each reasoning item carries a "summary" array of summary_text entries — these become trace.Steps.
trace.Signed is always false for OpenAI: the API returns the summary in plaintext and accepts unsigned mutations on round-trip. Anthropic differs; see the Anthropic adapter for the signed-trace path.
Empty-trace handling: o3 omits the reasoning summary >90% of the time per community reports. The bridge surfaces an empty trace.Steps; H-CoT then emits SkipReasoningTraceEmpty after exhausting its retry budget.
v0.10.0 #166 — OpenAI ImageProvider implementation.
The Chat Completions API accepts a multimodal content array (text + image_url parts) on user messages. Inline image bytes are base64-encoded as data: URLs; URL refs pass through verbatim. The Detail hint is forwarded to OpenAI as the image_url.detail field ("low" → fixed 85 tokens, "high" → tiled at 768px, "auto" → server decides).
This intentionally builds a separate request struct rather than reusing the text-only ChatCompletion path: the wire shape diverges (content as array vs string), and modeling them on one struct would force a pointer-y union that obscures the diff between the two paths.
Index ¶
- func NewOpenAIProvider(config *core.ProviderConfig) (core.Provider, error)
- type OpenAIProvider
- func (p *OpenAIProvider) ChatCompletion(ctx context.Context, request *core.ChatCompletionRequest) (*core.ChatCompletionResponse, error)
- func (p *OpenAIProvider) ChatWithImages(ctx context.Context, request *core.ChatCompletionRequest, ...) (*core.ChatCompletionResponse, error)
- func (p *OpenAIProvider) ChatWithReasoning(ctx context.Context, request *core.ChatCompletionRequest) (*core.ChatCompletionResponse, *core.ThinkingTrace, error)
- func (p *OpenAIProvider) Close() error
- func (p *OpenAIProvider) CountTokens(ctx context.Context, text string, modelID string) (int, error)
- func (p *OpenAIProvider) CreateEmbedding(ctx context.Context, request *core.EmbeddingRequest) (*core.EmbeddingResponse, error)
- func (p *OpenAIProvider) GetModels(ctx context.Context) ([]core.ModelInfo, error)
- func (p *OpenAIProvider) StreamingChatCompletion(ctx context.Context, request *core.ChatCompletionRequest, ...) error
- func (p *OpenAIProvider) TextCompletion(ctx context.Context, request *core.TextCompletionRequest) (*core.TextCompletionResponse, error)
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func NewOpenAIProvider ¶
func NewOpenAIProvider(config *core.ProviderConfig) (core.Provider, error)
NewOpenAIProvider creates a new OpenAI provider
Types ¶
type OpenAIProvider ¶
type OpenAIProvider struct {
*core.BaseProvider
// contains filtered or unexported fields
}
OpenAIProvider is an implementation of the Provider interface for OpenAI
func (*OpenAIProvider) ChatCompletion ¶
func (p *OpenAIProvider) ChatCompletion(ctx context.Context, request *core.ChatCompletionRequest) (*core.ChatCompletionResponse, error)
ChatCompletion generates a chat completion
func (*OpenAIProvider) ChatWithImages ¶ added in v0.10.0
func (p *OpenAIProvider) ChatWithImages(ctx context.Context, request *core.ChatCompletionRequest, images []core.ImageInput) (*core.ChatCompletionResponse, error)
ChatWithImages implements core.ImageProvider.
The supplied images are appended to the LAST user message in the request; if no user message is present a synthetic empty-text user message is appended to host the image parts. Existing system / assistant messages pass through as plain text content parts so the wire shape stays uniformly multimodal.
func (*OpenAIProvider) ChatWithReasoning ¶ added in v0.10.0
func (p *OpenAIProvider) ChatWithReasoning(ctx context.Context, request *core.ChatCompletionRequest) (*core.ChatCompletionResponse, *core.ThinkingTrace, error)
ChatWithReasoning implements core.ReasoningProvider via the OpenAI Responses API. The supplied request's Messages map to the input array; MaxTokens / Temperature pass through. Reasoning summaries are extracted into the returned ThinkingTrace; the final assistant text becomes the ChatCompletionResponse's first choice.
func (*OpenAIProvider) Close ¶
func (p *OpenAIProvider) Close() error
Close closes the provider and releases any resources
func (*OpenAIProvider) CountTokens ¶
CountTokens counts the number of tokens in a text
func (*OpenAIProvider) CreateEmbedding ¶
func (p *OpenAIProvider) CreateEmbedding(ctx context.Context, request *core.EmbeddingRequest) (*core.EmbeddingResponse, error)
CreateEmbedding creates an embedding
func (*OpenAIProvider) StreamingChatCompletion ¶
func (p *OpenAIProvider) StreamingChatCompletion(ctx context.Context, request *core.ChatCompletionRequest, callback func(response *core.ChatCompletionResponse) error) error
StreamingChatCompletion generates a streaming chat completion
func (*OpenAIProvider) TextCompletion ¶
func (p *OpenAIProvider) TextCompletion(ctx context.Context, request *core.TextCompletionRequest) (*core.TextCompletionResponse, error)
TextCompletion generates a text completion
Source Files