Documentation
¶
Index ¶
- Constants
- Variables
- type Client
- func (c *Client) Call(ctx context.Context, messageRequest types.CompletionRequest, env []string, ...) (*types.CompletionMessage, error)
- func (c *Client) ListModels(ctx context.Context, providers ...string) ([]openai.Model, error)
- func (c *Client) ProxyInfo([]string) (token, urlBase string)
- func (c *Client) RetrieveAPIKey(ctx context.Context, env []string) error
- func (c *Client) Supports(ctx context.Context, modelName string) (bool, error)
- func (c *Client) ValidAuth() error
- type InvalidAuthError
- type Options
Constants ¶
const ( DefaultModel = openai.GPT4o BuiltinCredName = "sys.openai" TooLongMessage = "Error: tool call output is too long" )
const DefaultMaxTokens = 400_000 // This is the limit for GPT-5
const DefaultTemperature float32 = 0.7
DefaultTemperature is sent when a tool sets no temperature. (Anthropic accepts only 0-1; 0.7 is inside that too.) It used to be 0. Reasoning models do badly under greedy decoding: replaying a bpai request against gpt-oss-120b, temperature 0 looped in the reasoning until the provider's length cap and returned neither content nor a tool call in every run, while 0.7 and 1.0 answered in 10 of 10. Leaving the field out is not enough either: some backends then decode greedily themselves (2 of 8 unset runs stalled the same way). 0.7 is Qwen's recommended setting, a common chat default, and inside gpt-oss's healthy range.
const WaitingMessage = "Waiting for model response..."
Variables ¶
var ErrNoVisibleOutput = errors.New("the model produced no answer in time")
ErrNoVisibleOutput is returned when a stream keeps delivering data but no text or tool call for longer than the first-output timeout. Reasoning deltas are not part of the answer, so a model caught in a reasoning loop looks alive to the idle timeout while the user sees nothing.
var ErrStreamIdle = errors.New("the model stopped sending output; the response is incomplete")
ErrStreamIdle is returned when a streamed completion delivers no data for longer than the idle timeout. Without it a silent upstream holds the run until the run-level timeout.
Functions ¶
This section is empty.
Types ¶
type Client ¶
type Client struct {
// contains filtered or unexported fields
}
func NewClient ¶
func NewClient(ctx context.Context, credStore credentials.CredentialStore, opts ...Options) (*Client, error)
func (*Client) Call ¶
func (c *Client) Call(ctx context.Context, messageRequest types.CompletionRequest, env []string, status chan<- types.CompletionStatus) (*types.CompletionMessage, error)
func (*Client) ListModels ¶
func (*Client) RetrieveAPIKey ¶
type InvalidAuthError ¶
type InvalidAuthError struct{}
func (InvalidAuthError) Error ¶
func (InvalidAuthError) Error() string
type Options ¶
type Options struct {
BaseURL string `usage:"OpenAI base URL" name:"openai-base-url" env:"OPENAI_BASE_URL"`
APIKey string `usage:"OpenAI API KEY" name:"openai-api-key" env:"OPENAI_API_KEY"`
OrgID string `usage:"OpenAI organization ID" name:"openai-org-id" env:"OPENAI_ORG_ID"`
DefaultModel string `usage:"Default LLM model to use" default:"gpt-4o"`
ConfigFile string `usage:"Path to GPTScript config file" name:"config"`
SetSeed bool `usage:"-"`
CacheKey string `usage:"-"`
Cache *cache.Client
}
Source Files