Documentation
¶
Index ¶
- type Analyzer
- func (a *Analyzer) AnalyzeBase64(ctx context.Context, b64data string, question string) (string, error)
- func (a *Analyzer) AnalyzeBase64WithConfidence(ctx context.Context, b64data, question string, threshold confidence.Score) (Result, error)
- func (a *Analyzer) AnalyzeFile(ctx context.Context, path string, question string) (string, error)
- func (a *Analyzer) AnalyzeFileWithConfidence(ctx context.Context, path, question string, threshold confidence.Score) (Result, error)
- type Result
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type Analyzer ¶
type Analyzer struct {
// contains filtered or unexported fields
}
func (*Analyzer) AnalyzeBase64 ¶
func (*Analyzer) AnalyzeBase64WithConfidence ¶ added in v0.54.0
func (a *Analyzer) AnalyzeBase64WithConfidence(ctx context.Context, b64data, question string, threshold confidence.Score) (Result, error)
AnalyzeBase64WithConfidence mirrors AnalyzeFileWithConfidence for data-URL / raw-base64 inputs.
func (*Analyzer) AnalyzeFile ¶
AnalyzeFile remains the legacy string-only entry point for callers that don't care about confidence routing. New callers should prefer AnalyzeFileWithConfidence so they can route low-confidence results to a clarifying user-facing question.
func (*Analyzer) AnalyzeFileWithConfidence ¶ added in v0.54.0
func (a *Analyzer) AnalyzeFileWithConfidence(ctx context.Context, path, question string, threshold confidence.Score) (Result, error)
AnalyzeFileWithConfidence is the P3-29 entrypoint. threshold is the per-call abstention threshold; pass 0 to use confidence.DefaultClassifierThreshold. Behaviour matches AnalyzeFile when the model doesn't emit a confidence signal — Result.Text is populated with the original prose and LowConfidence is false.
type Result ¶ added in v0.54.0
type Result struct {
Text string
Confidence confidence.Score
HasConfidence bool
LowConfidence bool
}
Result is the typed envelope returned by AnalyzeWithConfidence (roadmap P3-29). Text is the model's natural-language identification of the device / remote / tag / label in the image. Confidence is the model's self-graded certainty in [0, 1]; HasConfidence reports whether the model actually emitted a confidence value (a `false` here means "no signal — treat Confidence as the default 1.0 only for the purpose of comparing against thresholds, not as a true claim of certainty"). LowConfidence is precomputed against the caller-supplied threshold so the operator-facing renderer can flip to a clarifying-question UX.
Source Files