@doclo/core 0.1.11 → 0.1.12

package/dist/index.d.ts

@@ -1,5 +1,5 @@
-import { P as ProviderVendor, A as AccessMethod } from './validation-D_EcHqPl.js';
-export { G as AggregatedMetrics, B as BBox, r as CategorizeNodeConfig, t as ChunkMetadata, v as ChunkNodeConfig, u as ChunkOutput, n as CitationConfig, k as CitationSourceType, w as CombineNodeConfig, U as CompatibilityRule, C as ConsensusConfig, e as ConsensusMetadata, d as ConsensusRunResult, D as DocumentIR, b as DocumentIRExtras, y as EnhancedExtractionSchema, s as ExtractInputMode, E as ExtractNodeConfig, a0 as ExtractedImage, m as FieldCitation, F as FieldVotingDetails, H as FlowContext, a8 as FlowExecutionError, h as FlowInput, i as FlowInputValidation, j as FlowResult, a4 as FlowStepLocation, a9 as FlowValidationError, I as IRLine, a as IRPage, X as JSONSchemaNode, c as LLMJsonProvider, L as LLMProvider, _ as LanguageOptions, l as LineCitation, g as MaybeWithConsensusMetadata, M as MultimodalInput, aa as NODE_COMPATIBILITY_MATRIX, J as NodeCtx, Q as NodeDef, K as NodeTypeInfo, T as NodeTypeName, N as NormalizedBBox, O as OCRProvider, a1 as OCRProviderOptions, x as OutputNodeConfig, o as OutputWithCitations, f as OutputWithConsensus, Z as PageRangeOptions, p as ParseNodeConfig, Y as ProcessingMode, a3 as ProviderCitation, am as ProviderIdentity, aj as RESERVED_VARIABLES, R as ReasoningConfig, $ as SegmentationResult, S as SplitDocument, q as SplitNodeConfig, z as StepMetric, V as VLMProvider, a2 as VLMProviderOptions, W as ValidationResult, a5 as aggregateMetrics, ah as canStartForEachItemFlow, aq as createIdentity, al as extractErrorMessage, ad as getCompatibleTargets, ac as getNodeTypeInfo, ab as getNodeTypeName, ae as getSuggestedConnections, ag as getValidForEachStarters, ap as isLocalEndpoint, a6 as node, ao as parseProviderString, ak as protectReservedVariables, a7 as runPipeline, an as toProviderString, ai as validateJson, af as validateNodeConnection } from './validation-D_EcHqPl.js';
+import { P as ProviderVendor, A as AccessMethod } from './validation-wlK06puw.js';
+export { G as AggregatedMetrics, B as BBox, r as CategorizeNodeConfig, t as ChunkMetadata, v as ChunkNodeConfig, u as ChunkOutput, n as CitationConfig, k as CitationSourceType, w as CombineNodeConfig, U as CompatibilityRule, C as ConsensusConfig, e as ConsensusMetadata, d as ConsensusRunResult, D as DocumentIR, b as DocumentIRExtras, y as EnhancedExtractionSchema, s as ExtractInputMode, E as ExtractNodeConfig, a0 as ExtractedImage, m as FieldCitation, F as FieldVotingDetails, H as FlowContext, a8 as FlowExecutionError, h as FlowInput, i as FlowInputValidation, j as FlowResult, a4 as FlowStepLocation, a9 as FlowValidationError, I as IRLine, a as IRPage, X as JSONSchemaNode, c as LLMJsonProvider, L as LLMProvider, _ as LanguageOptions, l as LineCitation, g as MaybeWithConsensusMetadata, M as MultimodalInput, aa as NODE_COMPATIBILITY_MATRIX, J as NodeCtx, Q as NodeDef, K as NodeTypeInfo, T as NodeTypeName, N as NormalizedBBox, O as OCRProvider, a1 as OCRProviderOptions, x as OutputNodeConfig, o as OutputWithCitations, f as OutputWithConsensus, Z as PageRangeOptions, p as ParseNodeConfig, Y as ProcessingMode, a3 as ProviderCitation, am as ProviderIdentity, aj as RESERVED_VARIABLES, R as ReasoningConfig, $ as SegmentationResult, S as SplitDocument, q as SplitNodeConfig, z as StepMetric, V as VLMProvider, a2 as VLMProviderOptions, W as ValidationResult, a5 as aggregateMetrics, ah as canStartForEachItemFlow, aq as createIdentity, al as extractErrorMessage, ad as getCompatibleTargets, ac as getNodeTypeInfo, ab as getNodeTypeName, ae as getSuggestedConnections, ag as getValidForEachStarters, ap as isLocalEndpoint, a6 as node, ao as parseProviderString, ak as protectReservedVariables, a7 as runPipeline, an as toProviderString, ai as validateJson, af as validateNodeConnection } from './validation-wlK06puw.js';
 export { getDocumentPageCount, getPDFPageCount, getPageCountMetadata, getTotalPageCount, splitPDFIntoChunks } from './pdf-utils.js';
 
 /**
@@ -559,6 +559,8 @@ type NormalizedFeatures = {
     schemaValidation: boolean;
     /** Handwritten text recognition support */
     handwrittenText: boolean;
+    /** Separate header/footer extraction from main content */
+    headerFooterExtraction: boolean;
     /** Supported output formats */
     outputFormats: OutputFormatSupport;
 };

package/dist/index.js

@@ -1558,6 +1558,8 @@ function defaultNormalizer(id, data, source) {
     return normalizeReductoProvider(id, d);
   } else if (source === "unsiloed") {
     return normalizeUnsiloedProvider(id, d);
+  } else if (source === "mistral") {
+    return normalizeMistralProvider(id, d);
   }
   const defaultOutputFormats = { text: true, markdown: false, html: false, json: false };
   const defaultFeatures = {
@@ -1581,6 +1583,7 @@ function defaultNormalizer(id, data, source) {
     boundingBoxes: false,
     schemaValidation: false,
     handwrittenText: false,
+    headerFooterExtraction: false,
     outputFormats: defaultOutputFormats
   };
   return {
@@ -1670,6 +1673,8 @@ function normalizeLLMProvider(id, d) {
     // Some LLMs support schema validation
     handwrittenText: false,
     // Not specific to LLMs
+    headerFooterExtraction: false,
+    // LLMs don't extract header/footer separately
     outputFormats
   };
   const vendor = d.vendor ?? id;
@@ -1777,6 +1782,8 @@ function normalizeDatalabProvider(id, d) {
     // VLM providers support schema validation
     handwrittenText: true,
     // Datalab handles handwritten text
+    headerFooterExtraction: false,
+    // Datalab has issues with header/footer extraction
     outputFormats
   };
   return {
@@ -1886,6 +1893,8 @@ function normalizeReductoProvider(id, d) {
     // Extract has schema validation
     handwrittenText: false,
     // Reducto doesn't specifically advertise handwriting
+    headerFooterExtraction: true,
+    // Reducto has Header/Footer block types
     outputFormats
   };
   return {
@@ -1999,6 +2008,8 @@ function normalizeUnsiloedProvider(id, d) {
     // Extract supports schema validation
     handwrittenText: false,
     // Unsiloed doesn't specifically advertise handwriting
+    headerFooterExtraction: false,
+    // Unsiloed doesn't extract header/footer separately
     outputFormats
   };
   return {
@@ -2288,6 +2299,123 @@ function getAllModels() {
 function clearModelRegistry() {
   modelRegistry.clear();
 }
+function normalizeMistralProvider(id, d) {
+  const opts = d.supportedOptions ?? {};
+  const isVLM = d.type === "VLM";
+  const isOCR = d.type === "OCR";
+  const model = d.model ?? id;
+  const outputFormats = {
+    text: true,
+    markdown: d.outputFormat?.features?.markdown ?? isOCR,
+    html: d.outputFormat?.features?.htmlTables ?? isOCR,
+    // OCR 3 can output HTML tables
+    json: d.outputFormat?.features?.structuredJSON ?? isVLM
+  };
+  const features = {
+    maxPages: d.inputFormats?.maxPages !== void 0,
+    pageRange: true,
+    // Mistral supports pages param: "0-5" or [0,2,5]
+    languageHints: false,
+    // Mistral doesn't support language hints
+    processingModes: false,
+    // Mistral doesn't have processing modes
+    agenticMode: false,
+    // Mistral doesn't have agentic mode
+    customPrompts: false,
+    // Mistral OCR 3 doesn't support custom prompts
+    imageExtraction: opts.includeImageBase64 ?? false,
+    // Can include embedded images
+    pageMarkers: false,
+    // Mistral doesn't add page markers
+    citations: false,
+    // Mistral doesn't provide citations
+    chunking: false,
+    // Mistral doesn't do chunking
+    segmentation: false,
+    // Mistral doesn't do segmentation
+    stripExistingOCR: false,
+    formatLines: false,
+    forceOCR: true,
+    // OCR 3 always does OCR
+    tableOutputFormats: opts.tableFormat ?? isOCR,
+    // html or markdown table format
+    tableMerging: false,
+    confidence: false,
+    // Mistral doesn't provide confidence scores
+    boundingBoxes: d.outputFormat?.features?.boundingBoxes ?? false,
+    // NO text-level bboxes
+    schemaValidation: d.outputFormat?.features?.schemaValidation ?? isVLM,
+    // VLM supports schema
+    handwrittenText: d.outputFormat?.features?.handwrittenText ?? true,
+    // Excellent handwriting support
+    headerFooterExtraction: opts.extractHeader ?? opts.extractFooter ?? false,
+    // extract_header/extract_footer
+    outputFormats
+  };
+  return {
+    id: d.id ?? id,
+    name: d.name ?? id,
+    source: "mistral",
+    type: d.type ?? "OCR",
+    // 3-layer identity
+    identity: {
+      provider: "mistral",
+      model,
+      method: "native"
+    },
+    capabilities: {
+      supportsImages: d.capabilities?.supportsImages ?? true,
+      supportsPDFs: d.capabilities?.supportsPDFs ?? true,
+      supportsDocuments: d.capabilities?.supportsDocuments ?? false,
+      // DOCX/PPTX has known issues
+      supportsReasoning: false,
+      // OCR 3 doesn't do reasoning
+      supportsStructuredOutput: d.capabilities?.supportsStructuredOutput ?? isVLM,
+      // Extended capabilities
+      supportsPrompts: false,
+      supportsCitations: false,
+      supportsChunking: false,
+      supportsImageExtraction: opts.includeImageBase64 ?? false,
+      supportsPageMarkers: false,
+      supportsLanguageHints: false,
+      supportsProcessingModes: false,
+      supportsSegmentation: false,
+      outputFormats
+    },
+    features,
+    // Mistral providers always need raw document input
+    inputRequirements: {
+      inputType: d.inputRequirements?.inputType ?? "raw-document",
+      acceptedMethods: d.inputRequirements?.acceptedMethods ?? ["base64", "url"]
+    },
+    compatibleNodes: {
+      parse: d.compatibleNodes?.parse ?? isOCR,
+      extract: d.compatibleNodes?.extract ?? isVLM,
+      categorize: d.compatibleNodes?.categorize ?? false,
+      qualify: d.compatibleNodes?.qualify ?? false,
+      split: d.compatibleNodes?.split ?? false
+    },
+    inputFormats: {
+      imageMimeTypes: (d.inputFormats?.mimeTypes ?? []).filter((m) => m.startsWith("image/")),
+      documentMimeTypes: (d.inputFormats?.mimeTypes ?? []).filter((m) => !m.startsWith("image/")),
+      inputMethods: d.inputFormats?.inputMethods ?? ["base64", "url"],
+      maxFileSize: d.inputFormats?.maxFileSize ?? 50,
+      // 50MB limit
+      maxPages: d.inputFormats?.maxPages ?? 1e3
+    },
+    pricing: {
+      model: "per-page",
+      perPage: d.pricing?.perPage ?? 2e-3,
+      // $2/1000 pages
+      currency: "USD",
+      notes: d.pricing?.notes ?? "$2 per 1000 pages"
+    },
+    rateLimits: {
+      docsPerMinute: d.apiConfig?.rateLimit?.docsPerMinute
+    },
+    raw: d
+  };
+}
 
 // src/retry.ts
 var DEFAULT_RETRY_CONFIG = {