@canarycoders/ai 0.6.0 → 0.7.0

package/dist/index.d.cts

@@ -989,6 +989,112 @@ declare class VisionResource extends BaseResource {
     }>;
 }
 
+type OcrRecognitionLevel = "fast" | "accurate";
+type OcrGranularity = "line" | "word";
+type RedactStyle = "blur" | "box" | "pixelate";
+type RedactPreset = "email" | "phone" | "iban" | "credit_card" | "dni_nie";
+interface OcrRecognizeParams {
+    /** base64-encoded image bytes (max ~50MB); PNG/JPEG/HEIC/TIFF/WebP */
+    image: string;
+    /** default "accurate" */
+    recognitionLevel?: OcrRecognitionLevel;
+    /** BCP-47 codes; omit for automatic language detection */
+    languages?: string[];
+    /** domain words that bias language correction */
+    customWords?: string[];
+    minConfidence?: number;
+    /** "word" adds per-word boxes; default "line" */
+    granularity?: OcrGranularity;
+    tag?: string;
+    service?: string;
+}
+interface OcrWord {
+    text: string;
+    confidence: number;
+    bbox: BoundingBox;
+}
+interface OcrLine {
+    text: string;
+    confidence: number;
+    bbox: BoundingBox;
+    /** present at word granularity */
+    words?: OcrWord[];
+}
+interface OcrResult {
+    /** all lines joined with \n */
+    text: string;
+    lines: OcrLine[];
+    granularity: string;
+    model: string;
+    provider: string;
+    requestId: string;
+    inferenceTimeMs: number;
+    imageSize: {
+        width: number;
+        height: number;
+    };
+}
+interface RedactParams {
+    /** base64-encoded image bytes (max ~50MB) */
+    image: string;
+    /** explicit pixel regions (top-left origin) to redact without OCR */
+    regions?: BoundingBox[];
+    /** regex patterns matched against recognized text */
+    patterns?: string[];
+    /** built-in PII patterns */
+    presets?: RedactPreset[];
+    /** default "word": redact just the matched substring, whole line as fallback */
+    granularity?: OcrGranularity;
+    /** default "blur"; "box" is the safest for hard PII redaction */
+    style?: RedactStyle;
+    blurStrength?: number;
+    /** hex fill for box style, e.g. "#000000" */
+    fillColor?: string;
+    /** extra pixels around each redacted rect (default 4) */
+    padding?: number;
+    recognitionLevel?: OcrRecognitionLevel;
+    languages?: string[];
+    customWords?: string[];
+    minConfidence?: number;
+    tag?: string;
+    service?: string;
+}
+interface RedactionDetection {
+    /** matched text (absent for explicit regions) */
+    text?: string;
+    /** preset name, the regex source, or "explicit_region" */
+    pattern: string;
+    bbox: BoundingBox;
+    lineIndex?: number;
+    granularity: string;
+}
+interface RedactResult {
+    /** base64 PNG with redactions applied */
+    redactedImage: string;
+    detections: RedactionDetection[];
+    regionsRedacted: number;
+    counts: Record<string, number>;
+    model: string;
+    provider: string;
+    requestId: string;
+    inferenceTimeMs: number;
+    imageSize: {
+        width: number;
+        height: number;
+    };
+}
+interface OcrLanguages {
+    recognitionLevels: Record<string, string[]>;
+}
+
+declare class OcrResource extends BaseResource {
+    recognize(params: OcrRecognizeParams, poll?: PollOptions): Promise<OcrResult>;
+    recognizeJob(params: OcrRecognizeParams, signal?: AbortSignal): Promise<Job<OcrResult>>;
+    redact(params: RedactParams, poll?: PollOptions): Promise<RedactResult>;
+    redactJob(params: RedactParams, signal?: AbortSignal): Promise<Job<RedactResult>>;
+    languages(signal?: AbortSignal): Promise<OcrLanguages>;
+}
+
 interface CanaryCodersAIOptions {
     /** API key. Defaults to `process.env.CANARY_AI_API_KEY` (legacy `CANARYLLM_API_KEY` still read). */
     apiKey?: string;
@@ -1016,6 +1122,7 @@ declare class CanaryCodersAI {
     readonly audio: AudioResource;
     readonly embeddings: EmbeddingsResource;
     readonly vision: VisionResource;
+    readonly ocr: OcrResource;
     readonly conversations: ConversationsResource;
     readonly agents: AgentsResource;
     readonly realtime: RealtimeResource;
@@ -1100,4 +1207,4 @@ interface StreamAdapterOptions extends SSEOptions {
 
 // @ts-ignore
 export = CanaryCodersAI;
-export { APIConnectionError, APIConnectionTimeoutError, APIError, type ApiEnvelope, type AudioOutputFormat, AuthenticationError, type AutoLabelParams, type AutoLabelResult, type AutoTrainParams, BadRequestError, type BoundingBox, BrokeredCredential, CanaryCodersAI, type CanaryCodersAIOptions, CanaryCodersAI as CanaryLLM, type CanaryCodersAIOptions as CanaryLLMOptions, type ChatProvider, type ChatStreamEvent, type ChatStreamOptions, type Citation, CompatTarget, type CompleteParams, type CompletionResult, ConflictError, type ContentPart, type ConversationQuestion, type ConversationSession, type ConversationSessionRecord, type ConversationTemplate, type ConversationTemplateParams, type ConversationTemplateType, type ConversationTemplateUpdate, type ConversationTool, type ConversationVoiceSettings, type CreateConversationSessionParams, CreateRealtimeSessionParams, type DetectParams, type Detection, type DialogueInput, type DialogueParams, type DialogueResult, type DocumentContent, type EmbeddingParams, type EmbeddingResult, type FaceDetectParams, type FaceDetectionResult, FinalizeRealtimeParams, type GeneratedImage, type GeneratedVideo, type ImageContent, type ImageGenerateParams, type ImageGenerationResult, type ImageRecognitionResult, InternalServerError, Job, type JobSnapshot, type JobStatus, type JobStreamOptions, type KeyInfo, type KeyValidation, type Message, type MessageContent, type MessageRole, type ModelCapability, type ModelInfo, type MusicParams, type MusicResult, NotFoundError, PermissionError, type PollOptions, type PortalPeriod, type ProviderType, RateLimitError, RealtimeSession, RealtimeSessionRecord, type ResponseFormat, type STTResult, type SignedUrlParams, type SignedUrlResult, type SoundEffectParams, type SoundEffectResult, type SpeechParams, type StreamAdapterOptions, type StreamChunk, type StreamProtocol, type TTSResult, type TaskKind, type TextContent, type ThinkingMode, type TokenUsage, type ToolCall, type ToolCallDelta, type ToolChoice, type ToolDefinition, type TrainParams, type TrainingJob, type TranscribeParams, type TranscriptWord, UnprocessableEntityError, type UsageSummary, type VideoContent, type VideoGenerateParams, type VideoGenerationResult, type VideoUploadOptions, type VideoUploadResult, type VisionModel, type VoiceInfo, type VoiceSettings, type WebDetectParams, type WebDetectionResult, type WebEntity, type WebSearchOptions, type ZeroShotDetectParams, toBrokeredCredential };
+export { APIConnectionError, APIConnectionTimeoutError, APIError, type ApiEnvelope, type AudioOutputFormat, AuthenticationError, type AutoLabelParams, type AutoLabelResult, type AutoTrainParams, BadRequestError, type BoundingBox, BrokeredCredential, CanaryCodersAI, type CanaryCodersAIOptions, CanaryCodersAI as CanaryLLM, type CanaryCodersAIOptions as CanaryLLMOptions, type ChatProvider, type ChatStreamEvent, type ChatStreamOptions, type Citation, CompatTarget, type CompleteParams, type CompletionResult, ConflictError, type ContentPart, type ConversationQuestion, type ConversationSession, type ConversationSessionRecord, type ConversationTemplate, type ConversationTemplateParams, type ConversationTemplateType, type ConversationTemplateUpdate, type ConversationTool, type ConversationVoiceSettings, type CreateConversationSessionParams, CreateRealtimeSessionParams, type DetectParams, type Detection, type DialogueInput, type DialogueParams, type DialogueResult, type DocumentContent, type EmbeddingParams, type EmbeddingResult, type FaceDetectParams, type FaceDetectionResult, FinalizeRealtimeParams, type GeneratedImage, type GeneratedVideo, type ImageContent, type ImageGenerateParams, type ImageGenerationResult, type ImageRecognitionResult, InternalServerError, Job, type JobSnapshot, type JobStatus, type JobStreamOptions, type KeyInfo, type KeyValidation, type Message, type MessageContent, type MessageRole, type ModelCapability, type ModelInfo, type MusicParams, type MusicResult, NotFoundError, type OcrGranularity, type OcrLanguages, type OcrLine, type OcrRecognitionLevel, type OcrRecognizeParams, type OcrResult, type OcrWord, PermissionError, type PollOptions, type PortalPeriod, type ProviderType, RateLimitError, RealtimeSession, RealtimeSessionRecord, type RedactParams, type RedactPreset, type RedactResult, type RedactStyle, type RedactionDetection, type ResponseFormat, type STTResult, type SignedUrlParams, type SignedUrlResult, type SoundEffectParams, type SoundEffectResult, type SpeechParams, type StreamAdapterOptions, type StreamChunk, type StreamProtocol, type TTSResult, type TaskKind, type TextContent, type ThinkingMode, type TokenUsage, type ToolCall, type ToolCallDelta, type ToolChoice, type ToolDefinition, type TrainParams, type TrainingJob, type TranscribeParams, type TranscriptWord, UnprocessableEntityError, type UsageSummary, type VideoContent, type VideoGenerateParams, type VideoGenerationResult, type VideoUploadOptions, type VideoUploadResult, type VisionModel, type VoiceInfo, type VoiceSettings, type WebDetectParams, type WebDetectionResult, type WebEntity, type WebSearchOptions, type ZeroShotDetectParams, toBrokeredCredential };

package/dist/index.d.ts

@@ -989,6 +989,112 @@ declare class VisionResource extends BaseResource {
     }>;
 }
 
+type OcrRecognitionLevel = "fast" | "accurate";
+type OcrGranularity = "line" | "word";
+type RedactStyle = "blur" | "box" | "pixelate";
+type RedactPreset = "email" | "phone" | "iban" | "credit_card" | "dni_nie";
+interface OcrRecognizeParams {
+    /** base64-encoded image bytes (max ~50MB); PNG/JPEG/HEIC/TIFF/WebP */
+    image: string;
+    /** default "accurate" */
+    recognitionLevel?: OcrRecognitionLevel;
+    /** BCP-47 codes; omit for automatic language detection */
+    languages?: string[];
+    /** domain words that bias language correction */
+    customWords?: string[];
+    minConfidence?: number;
+    /** "word" adds per-word boxes; default "line" */
+    granularity?: OcrGranularity;
+    tag?: string;
+    service?: string;
+}
+interface OcrWord {
+    text: string;
+    confidence: number;
+    bbox: BoundingBox;
+}
+interface OcrLine {
+    text: string;
+    confidence: number;
+    bbox: BoundingBox;
+    /** present at word granularity */
+    words?: OcrWord[];
+}
+interface OcrResult {
+    /** all lines joined with \n */
+    text: string;
+    lines: OcrLine[];
+    granularity: string;
+    model: string;
+    provider: string;
+    requestId: string;
+    inferenceTimeMs: number;
+    imageSize: {
+        width: number;
+        height: number;
+    };
+}
+interface RedactParams {
+    /** base64-encoded image bytes (max ~50MB) */
+    image: string;
+    /** explicit pixel regions (top-left origin) to redact without OCR */
+    regions?: BoundingBox[];
+    /** regex patterns matched against recognized text */
+    patterns?: string[];
+    /** built-in PII patterns */
+    presets?: RedactPreset[];
+    /** default "word": redact just the matched substring, whole line as fallback */
+    granularity?: OcrGranularity;
+    /** default "blur"; "box" is the safest for hard PII redaction */
+    style?: RedactStyle;
+    blurStrength?: number;
+    /** hex fill for box style, e.g. "#000000" */
+    fillColor?: string;
+    /** extra pixels around each redacted rect (default 4) */
+    padding?: number;
+    recognitionLevel?: OcrRecognitionLevel;
+    languages?: string[];
+    customWords?: string[];
+    minConfidence?: number;
+    tag?: string;
+    service?: string;
+}
+interface RedactionDetection {
+    /** matched text (absent for explicit regions) */
+    text?: string;
+    /** preset name, the regex source, or "explicit_region" */
+    pattern: string;
+    bbox: BoundingBox;
+    lineIndex?: number;
+    granularity: string;
+}
+interface RedactResult {
+    /** base64 PNG with redactions applied */
+    redactedImage: string;
+    detections: RedactionDetection[];
+    regionsRedacted: number;
+    counts: Record<string, number>;
+    model: string;
+    provider: string;
+    requestId: string;
+    inferenceTimeMs: number;
+    imageSize: {
+        width: number;
+        height: number;
+    };
+}
+interface OcrLanguages {
+    recognitionLevels: Record<string, string[]>;
+}
+
+declare class OcrResource extends BaseResource {
+    recognize(params: OcrRecognizeParams, poll?: PollOptions): Promise<OcrResult>;
+    recognizeJob(params: OcrRecognizeParams, signal?: AbortSignal): Promise<Job<OcrResult>>;
+    redact(params: RedactParams, poll?: PollOptions): Promise<RedactResult>;
+    redactJob(params: RedactParams, signal?: AbortSignal): Promise<Job<RedactResult>>;
+    languages(signal?: AbortSignal): Promise<OcrLanguages>;
+}
+
 interface CanaryCodersAIOptions {
     /** API key. Defaults to `process.env.CANARY_AI_API_KEY` (legacy `CANARYLLM_API_KEY` still read). */
     apiKey?: string;
@@ -1016,6 +1122,7 @@ declare class CanaryCodersAI {
     readonly audio: AudioResource;
     readonly embeddings: EmbeddingsResource;
     readonly vision: VisionResource;
+    readonly ocr: OcrResource;
     readonly conversations: ConversationsResource;
     readonly agents: AgentsResource;
     readonly realtime: RealtimeResource;
@@ -1098,4 +1205,4 @@ interface StreamAdapterOptions extends SSEOptions {
     includeRaw?: boolean;
 }
 
-export { APIConnectionError, APIConnectionTimeoutError, APIError, type ApiEnvelope, type AudioOutputFormat, AuthenticationError, type AutoLabelParams, type AutoLabelResult, type AutoTrainParams, BadRequestError, type BoundingBox, BrokeredCredential, CanaryCodersAI, type CanaryCodersAIOptions, CanaryCodersAI as CanaryLLM, type CanaryCodersAIOptions as CanaryLLMOptions, type ChatProvider, type ChatStreamEvent, type ChatStreamOptions, type Citation, CompatTarget, type CompleteParams, type CompletionResult, ConflictError, type ContentPart, type ConversationQuestion, type ConversationSession, type ConversationSessionRecord, type ConversationTemplate, type ConversationTemplateParams, type ConversationTemplateType, type ConversationTemplateUpdate, type ConversationTool, type ConversationVoiceSettings, type CreateConversationSessionParams, CreateRealtimeSessionParams, type DetectParams, type Detection, type DialogueInput, type DialogueParams, type DialogueResult, type DocumentContent, type EmbeddingParams, type EmbeddingResult, type FaceDetectParams, type FaceDetectionResult, FinalizeRealtimeParams, type GeneratedImage, type GeneratedVideo, type ImageContent, type ImageGenerateParams, type ImageGenerationResult, type ImageRecognitionResult, InternalServerError, Job, type JobSnapshot, type JobStatus, type JobStreamOptions, type KeyInfo, type KeyValidation, type Message, type MessageContent, type MessageRole, type ModelCapability, type ModelInfo, type MusicParams, type MusicResult, NotFoundError, PermissionError, type PollOptions, type PortalPeriod, type ProviderType, RateLimitError, RealtimeSession, RealtimeSessionRecord, type ResponseFormat, type STTResult, type SignedUrlParams, type SignedUrlResult, type SoundEffectParams, type SoundEffectResult, type SpeechParams, type StreamAdapterOptions, type StreamChunk, type StreamProtocol, type TTSResult, type TaskKind, type TextContent, type ThinkingMode, type TokenUsage, type ToolCall, type ToolCallDelta, type ToolChoice, type ToolDefinition, type TrainParams, type TrainingJob, type TranscribeParams, type TranscriptWord, UnprocessableEntityError, type UsageSummary, type VideoContent, type VideoGenerateParams, type VideoGenerationResult, type VideoUploadOptions, type VideoUploadResult, type VisionModel, type VoiceInfo, type VoiceSettings, type WebDetectParams, type WebDetectionResult, type WebEntity, type WebSearchOptions, type ZeroShotDetectParams, CanaryCodersAI as default, toBrokeredCredential };
+export { APIConnectionError, APIConnectionTimeoutError, APIError, type ApiEnvelope, type AudioOutputFormat, AuthenticationError, type AutoLabelParams, type AutoLabelResult, type AutoTrainParams, BadRequestError, type BoundingBox, BrokeredCredential, CanaryCodersAI, type CanaryCodersAIOptions, CanaryCodersAI as CanaryLLM, type CanaryCodersAIOptions as CanaryLLMOptions, type ChatProvider, type ChatStreamEvent, type ChatStreamOptions, type Citation, CompatTarget, type CompleteParams, type CompletionResult, ConflictError, type ContentPart, type ConversationQuestion, type ConversationSession, type ConversationSessionRecord, type ConversationTemplate, type ConversationTemplateParams, type ConversationTemplateType, type ConversationTemplateUpdate, type ConversationTool, type ConversationVoiceSettings, type CreateConversationSessionParams, CreateRealtimeSessionParams, type DetectParams, type Detection, type DialogueInput, type DialogueParams, type DialogueResult, type DocumentContent, type EmbeddingParams, type EmbeddingResult, type FaceDetectParams, type FaceDetectionResult, FinalizeRealtimeParams, type GeneratedImage, type GeneratedVideo, type ImageContent, type ImageGenerateParams, type ImageGenerationResult, type ImageRecognitionResult, InternalServerError, Job, type JobSnapshot, type JobStatus, type JobStreamOptions, type KeyInfo, type KeyValidation, type Message, type MessageContent, type MessageRole, type ModelCapability, type ModelInfo, type MusicParams, type MusicResult, NotFoundError, type OcrGranularity, type OcrLanguages, type OcrLine, type OcrRecognitionLevel, type OcrRecognizeParams, type OcrResult, type OcrWord, PermissionError, type PollOptions, type PortalPeriod, type ProviderType, RateLimitError, RealtimeSession, RealtimeSessionRecord, type RedactParams, type RedactPreset, type RedactResult, type RedactStyle, type RedactionDetection, type ResponseFormat, type STTResult, type SignedUrlParams, type SignedUrlResult, type SoundEffectParams, type SoundEffectResult, type SpeechParams, type StreamAdapterOptions, type StreamChunk, type StreamProtocol, type TTSResult, type TaskKind, type TextContent, type ThinkingMode, type TokenUsage, type ToolCall, type ToolCallDelta, type ToolChoice, type ToolDefinition, type TrainParams, type TrainingJob, type TranscribeParams, type TranscriptWord, UnprocessableEntityError, type UsageSummary, type VideoContent, type VideoGenerateParams, type VideoGenerationResult, type VideoUploadOptions, type VideoUploadResult, type VisionModel, type VoiceInfo, type VoiceSettings, type WebDetectParams, type WebDetectionResult, type WebEntity, type WebSearchOptions, type ZeroShotDetectParams, CanaryCodersAI as default, toBrokeredCredential };

package/dist/index.js

@@ -1584,6 +1584,27 @@ var VisionResource = class extends BaseResource {
   }
 };
 
+// src/resources/ocr.ts
+var OcrResource = class extends BaseResource {
+  recognize(params, poll) {
+    return this.runQueued("/api/ocr/recognize", params, "vision", poll);
+  }
+  recognizeJob(params, signal) {
+    return this.submitQueued("/api/ocr/recognize", params, "vision", signal);
+  }
+  redact(params, poll) {
+    return this.runQueued("/api/ocr/redact", params, "vision", poll);
+  }
+  redactJob(params, signal) {
+    return this.submitQueued("/api/ocr/redact", params, "vision", signal);
+  }
+  async languages(signal) {
+    return this.transport.json("GET", "/api/ocr/languages", {
+      signal
+    });
+  }
+};
+
 // src/client.ts
 var DEFAULT_BASE_URL = "https://api.ai.canarycoders.es";
 function readEnv(key) {
@@ -1598,6 +1619,7 @@ var CanaryCodersAI = class {
   audio;
   embeddings;
   vision;
+  ocr;
   conversations;
   agents;
   realtime;
@@ -1630,6 +1652,7 @@ var CanaryCodersAI = class {
     this.audio = new AudioResource(this.transport, poll);
     this.embeddings = new EmbeddingsResource(this.transport, poll);
     this.vision = new VisionResource(this.transport, poll);
+    this.ocr = new OcrResource(this.transport, poll);
     this.conversations = new ConversationsResource(this.transport, poll);
     this.agents = new AgentsResource(this.transport, poll);
     this.realtime = new RealtimeResource(this.transport, poll);