Documentation
¶
Index ¶
- type ChatCompletion
- type Option
- type ReasoningEffort
- type Responses
- type ResponsesOption
- func WithResponsesReasoningEffort(effort ReasoningEffort) ResponsesOption
- func WithResponsesRetryConfig(cfg retry.Config) ResponsesOption
- func WithResponsesServiceTier(tier ServiceTier) ResponsesOption
- func WithResponsesStore(store bool) ResponsesOption
- func WithResponsesThinkingEnabled(enabled bool) ResponsesOption
- type ServiceTier
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type ChatCompletion ¶
type ChatCompletion struct {
// contains filtered or unexported fields
}
ChatCompletion implements model.LLM using the OpenAI Chat Completions API.
func New ¶
func New(apiKey, baseURL, modelName string, retryCfg ...retry.Config) *ChatCompletion
New creates a new ChatCompletion instance. apiKey is required. baseURL is optional; when non-empty it overrides the default OpenAI endpoint, which allows using any OpenAI-compatible provider. retryCfg is optional; when provided it enables automatic retry with exponential backoff on transient errors (rate limits, 5xx, network issues).
func NewWithOptions ¶ added in v0.0.7
func NewWithOptions(apiKey, baseURL, modelName string, opts ...Option) *ChatCompletion
NewWithOptions creates a new ChatCompletion instance with explicit provider compatibility, retry, and generation options.
func (*ChatCompletion) GenerateContent ¶
func (c *ChatCompletion) GenerateContent(ctx context.Context, req *model.LLMRequest, cfg *model.GenerateConfig, stream bool) iter.Seq2[*model.LLMResponse, error]
GenerateContent sends the request to the OpenAI Chat Completions API. When stream is false, exactly one complete *model.LLMResponse is yielded. When stream is true, partial text chunks are yielded (Partial=true) followed by the assembled complete response (Partial=false, TurnComplete=true). Transient errors are automatically retried according to the retry.Config provided at construction time.
func (*ChatCompletion) Name ¶
func (c *ChatCompletion) Name() string
Name returns the model identifier.
type Option ¶ added in v0.0.7
type Option func(*ChatCompletion)
Option configures a ChatCompletion instance.
func WithDeepSeekCompatibility ¶ added in v0.0.7
func WithDeepSeekCompatibility() Option
WithDeepSeekCompatibility configures ChatCompletion for DeepSeek's OpenAI-compatible Chat Completions API.
func WithReasoningEffort ¶ added in v0.0.7
func WithReasoningEffort(effort ReasoningEffort) Option
WithReasoningEffort sets OpenAI-compatible reasoning effort for every call.
func WithRetryConfig ¶ added in v0.0.7
WithRetryConfig sets the retry behavior for transient API errors.
func WithServiceTier ¶ added in v0.0.7
func WithServiceTier(tier ServiceTier) Option
WithServiceTier sets the OpenAI-compatible service tier for every call.
func WithThinkingEnabled ¶ added in v0.0.7
WithThinkingEnabled explicitly enables or disables provider-specific thinking controls for every call. A disabled value maps to reasoning_effort "none" for providers that do not expose a separate thinking toggle.
type ReasoningEffort ¶ added in v0.0.7
type ReasoningEffort string
ReasoningEffort controls OpenAI-compatible reasoning effort.
const ( // ReasoningEffortNone disables reasoning effort when supported by the provider. ReasoningEffortNone ReasoningEffort = "none" // ReasoningEffortMinimal requests minimal reasoning effort. ReasoningEffortMinimal ReasoningEffort = "minimal" // ReasoningEffortLow requests low reasoning effort. ReasoningEffortLow ReasoningEffort = "low" // ReasoningEffortMedium requests medium reasoning effort. ReasoningEffortMedium ReasoningEffort = "medium" // ReasoningEffortHigh requests high reasoning effort. ReasoningEffortHigh ReasoningEffort = "high" // ReasoningEffortXhigh requests extra-high reasoning effort on compatible providers. ReasoningEffortXhigh ReasoningEffort = "xhigh" )
type Responses ¶ added in v0.0.9
type Responses struct {
// contains filtered or unexported fields
}
Responses implements model.LLM using the OpenAI Responses API.
By default it sends the full ADK-provided history on every request with store=false, preserving ADK's own session ledger as the source of truth.
func NewResponses ¶ added in v0.0.9
NewResponses creates a new Responses instance. apiKey is required. baseURL is optional; when non-empty it overrides the default OpenAI endpoint. retryCfg is optional; when provided it enables automatic retry with exponential backoff on transient errors.
func NewResponsesWithOptions ¶ added in v0.0.9
func NewResponsesWithOptions(apiKey, baseURL, modelName string, opts ...ResponsesOption) *Responses
NewResponsesWithOptions creates a new Responses instance with explicit retry and generation options.
func (*Responses) GenerateContent ¶ added in v0.0.9
func (r *Responses) GenerateContent(ctx context.Context, req *model.LLMRequest, cfg *model.GenerateConfig, stream bool) iter.Seq2[*model.LLMResponse, error]
GenerateContent sends the request to the OpenAI Responses API. When stream is false, exactly one complete *model.LLMResponse is yielded. When stream is true, partial text chunks are yielded (Partial=true) followed by the assembled complete response (Partial=false, TurnComplete=true). Transient errors are automatically retried according to the retry.Config provided at construction time.
type ResponsesOption ¶ added in v0.0.9
type ResponsesOption func(*Responses)
ResponsesOption configures a Responses instance.
func WithResponsesReasoningEffort ¶ added in v0.0.9
func WithResponsesReasoningEffort(effort ReasoningEffort) ResponsesOption
WithResponsesReasoningEffort sets OpenAI Responses reasoning effort for every call.
func WithResponsesRetryConfig ¶ added in v0.0.9
func WithResponsesRetryConfig(cfg retry.Config) ResponsesOption
WithResponsesRetryConfig sets the retry behavior for transient API errors.
func WithResponsesServiceTier ¶ added in v0.0.9
func WithResponsesServiceTier(tier ServiceTier) ResponsesOption
WithResponsesServiceTier sets the OpenAI Responses service tier for every call.
func WithResponsesStore ¶ added in v0.0.9
func WithResponsesStore(store bool) ResponsesOption
WithResponsesStore controls whether OpenAI stores generated responses for later retrieval. The default is false so ADK remains the state owner.
func WithResponsesThinkingEnabled ¶ added in v0.0.9
func WithResponsesThinkingEnabled(enabled bool) ResponsesOption
WithResponsesThinkingEnabled controls OpenAI Responses reasoning for every call. Disabling maps to reasoning effort "none"; enabling leaves the effort level to the provider unless WithResponsesReasoningEffort is also set.
type ServiceTier ¶ added in v0.0.7
type ServiceTier string
ServiceTier specifies the OpenAI-compatible service tier for a request.
const ( // ServiceTierAuto lets the provider choose the service tier. ServiceTierAuto ServiceTier = "auto" // ServiceTierDefault uses the provider default service tier. ServiceTierDefault ServiceTier = "default" // ServiceTierFlex uses the flex service tier. ServiceTierFlex ServiceTier = "flex" // ServiceTierScale uses the scale service tier. ServiceTierScale ServiceTier = "scale" // ServiceTierPriority uses the priority service tier. ServiceTierPriority ServiceTier = "priority" )
Source Files