{"passport":{"unfragile":{"@version":"1.0","version":"2026-05","artifact":{"id":"scale-ai","slug":"scale-ai","name":"Scale AI","type":"platform","url":"https://scale.com","page_url":"https://unfragile.ai/scale-ai","categories":["model-training"],"tags":[],"pricing":{"model":"free","free":true,"starting_price":null},"status":"active","verified":false},"capabilities":[{"id":"scale-ai__cap_0","uri":"capability://data.processing.analysis.human.in.the.loop.image.annotation.with.quality.control","name":"human-in-the-loop image annotation with quality control","description":"Manages distributed annotation workflows for computer vision tasks (bounding boxes, segmentation, classification) through a managed workforce with built-in quality assurance layers. Uses consensus-based validation where multiple annotators label the same data and disagreements trigger expert review, combined with automated consistency checks and rework queues to maintain labeling accuracy above configurable thresholds.","intents":["I need to label thousands of images for object detection without hiring and managing annotators myself","I want to ensure annotation quality stays above 95% accuracy without manual spot-checking every label","I need to scale annotation from 100 to 100,000 images without changing my workflow or infrastructure"],"best_for":["autonomous vehicle teams building perception datasets","computer vision startups without in-house labeling infrastructure","enterprises requiring SOC 2 / FedRAMP compliant annotation workflows"],"limitations":["consensus-based QA adds 20-40% latency to annotation cycles compared to single-pass labeling","custom annotation schemas require JSON schema definition and may need 1-2 iteration cycles to optimize for workforce understanding","no real-time streaming annotation — batches must be submitted and processed asynchronously"],"requires":["image dataset in JPEG, PNG, or WebP format","annotation schema defined in Scale's JSON schema format or via web UI","API key for programmatic access (if using API rather than web dashboard)"],"input_types":["image files (JPEG, PNG, WebP, TIFF)","image URLs (HTTP/HTTPS)","video frames (extracted as images)","annotation schema (JSON)"],"output_types":["structured annotations (JSON with bounding boxes, polygons, keypoints, classifications)","confidence scores per annotation","annotator metadata and audit trails","quality metrics and rework flags"],"categories":["data-processing-analysis","human-in-the-loop"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"scale-ai__cap_1","uri":"capability://data.processing.analysis.nlp.text.annotation.and.entity.labeling.at.scale","name":"nlp text annotation and entity labeling at scale","description":"Handles sequence labeling, named entity recognition, intent classification, and semantic relationship annotation for text data through a managed annotation interface. Supports hierarchical entity schemas, multi-label classification, and context-aware labeling where annotators see surrounding text and previous labels to maintain consistency across large corpora.","intents":["I need to label 50,000 customer support tickets for intent classification and entity extraction without building an annotation tool","I want to maintain consistent entity tagging across a corpus where context matters (e.g., 'Apple' as company vs fruit)","I need to track which annotator labeled what and audit the labeling process for compliance"],"best_for":["NLP teams training intent classifiers and NER models for production","enterprises building domain-specific language models with labeled training data","government and regulated industries requiring full audit trails for data labeling"],"limitations":["hierarchical entity schemas with >50 entity types may cause annotator confusion and require extensive training","no built-in active learning — cannot automatically select most uncertain examples for labeling","turnaround time for large batches (10k+ examples) is 3-7 days depending on complexity and workforce availability"],"requires":["text data in plain text, CSV, or JSON format","entity schema or classification taxonomy defined upfront","minimum batch size of 100 examples for efficient processing"],"input_types":["plain text strings","CSV/JSON with text fields","pre-tokenized text with token boundaries","classification taxonomies (flat or hierarchical)"],"output_types":["token-level annotations (BIO/BIOES format)","entity spans with types and confidence","document-level classifications","relationship annotations (if schema supports)","annotator agreement metrics"],"categories":["data-processing-analysis","text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"scale-ai__cap_10","uri":"capability://data.processing.analysis.multi.language.annotation.support.with.native.speaker.workforce","name":"multi-language annotation support with native speaker workforce","description":"Provides annotation services in 50+ languages with native speaker annotators, supporting language-specific nuances, dialects, and cultural context. Automatically routes tasks to annotators matching required language and dialect, with quality assurance for language-specific tasks like machine translation evaluation and sentiment analysis across languages.","intents":["I need to label customer support data in 10 different languages but don't have native speakers on my team","I want to evaluate machine translation quality across multiple language pairs with native speaker judgment","I need to annotate sentiment and intent in regional dialects (e.g., Brazilian Portuguese vs European Portuguese)"],"best_for":["global companies building multilingual NLP models","machine translation companies evaluating translation quality","enterprises serving international markets with language-specific content moderation"],"limitations":["rare languages (< 1 million speakers) may have limited annotator availability and higher costs","dialect-specific annotation (e.g., Moroccan Arabic) requires specialized annotators and may have 2-4 week lead times","quality assurance for language-specific tasks is harder to automate; requires native speaker expert review"],"requires":["clear specification of required language and dialect","annotation schema adapted for language-specific considerations (e.g., grammatical structures, cultural context)","minimum batch size of 100 examples per language for efficient processing"],"input_types":["text in target language(s)","language and dialect specifications","annotation schema (adapted for language-specific nuances)"],"output_types":["language-specific annotations (with dialect metadata)","inter-annotator agreement per language","language-specific quality metrics"],"categories":["data-processing-analysis","text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"scale-ai__cap_11","uri":"capability://data.processing.analysis.model.assisted.annotation.with.pre.labeling.and.human.review","name":"model-assisted annotation with pre-labeling and human review","description":"Integrates with client ML models to pre-label data automatically, then routes pre-labeled data to human annotators for review and correction. Reduces annotation time by 40-60% compared to manual annotation from scratch by having annotators verify and correct model predictions rather than labeling from zero. Tracks which examples the model got wrong and uses those for model retraining.","intents":["I want to speed up annotation by having my model pre-label data and then have humans correct the mistakes","I want to identify which examples my model struggles with so I can retrain it on those hard cases","I want to reduce annotation costs by 50% without sacrificing quality"],"best_for":["teams with existing trained models who want to scale annotation efficiently","iterative model development workflows where annotation and retraining are tightly coupled","projects with large datasets where pre-labeling can provide significant cost savings"],"limitations":["pre-labeling quality depends on model accuracy; if model is <70% accurate, pre-labels may introduce more errors than they save","annotators may develop bias toward accepting model predictions (automation bias), reducing correction quality","requires model retraining pipeline to close the loop; pre-labeling alone doesn't improve model without retraining"],"requires":["trained ML model that can produce predictions on unlabeled data","model API or batch prediction capability","retraining pipeline to incorporate corrected labels back into model"],"input_types":["unlabeled data","model predictions (pre-labels)","model confidence scores (to prioritize uncertain predictions for review)"],"output_types":["corrected annotations (human-verified labels)","model error analysis (which predictions were wrong)","annotation time savings metrics"],"categories":["data-processing-analysis","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"scale-ai__cap_2","uri":"capability://data.processing.analysis.generative.ai.output.evaluation.and.rlhf.data.collection","name":"generative ai output evaluation and rlhf data collection","description":"Collects human feedback on LLM outputs (rankings, ratings, binary preferences) to create training data for reinforcement learning from human feedback (RLHF) and model fine-tuning. Manages comparison workflows where annotators rank multiple model outputs, rate quality on custom rubrics, or provide binary preference judgments, with built-in consistency checks and expert review for edge cases.","intents":["I need to collect preference data comparing outputs from GPT-4 and my fine-tuned model to train a reward model","I want to evaluate my LLM's responses on safety, factuality, and helpfulness using a custom rubric without building an evaluation interface","I need to scale RLHF data collection from 1,000 to 100,000 comparisons while maintaining annotator agreement above 85%"],"best_for":["LLM teams building reward models for RLHF fine-tuning","enterprises evaluating generative AI outputs for production deployment","research teams collecting human preference data for alignment research"],"limitations":["annotator fatigue on subjective tasks (e.g., rating helpfulness) can degrade agreement after 200+ examples per session","complex rubrics with >5 dimensions require extensive annotator training and may achieve only 70-75% agreement","no automatic detection of hallucinations or factual errors — requires annotators to verify claims independently"],"requires":["LLM outputs in text format (JSON or plain text)","evaluation rubric or preference schema defined upfront","minimum 100 examples for meaningful agreement statistics"],"input_types":["LLM outputs (text strings)","prompt-response pairs (JSON)","multiple model outputs for comparison","evaluation rubrics (custom scales and dimensions)"],"output_types":["preference rankings (e.g., Model A > Model B > Model C)","Likert-scale ratings (1-5 or custom range)","binary preference judgments","inter-annotator agreement scores (Fleiss' kappa, Krippendorff's alpha)","structured feedback and explanations"],"categories":["data-processing-analysis","safety-moderation"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"scale-ai__cap_3","uri":"capability://data.processing.analysis.autonomous.vehicle.perception.dataset.curation.and.versioning","name":"autonomous vehicle perception dataset curation and versioning","description":"Manages multi-modal sensor data (camera, LiDAR, radar) annotation and dataset versioning for autonomous vehicle training pipelines. Handles 3D bounding box annotation, sensor fusion labeling, and tracks dataset lineage with version control, allowing teams to reproduce model training runs and audit which data versions were used for which model checkpoints.","intents":["I need to annotate 3D bounding boxes on camera and LiDAR data for autonomous vehicle perception without building custom annotation tools","I want to version my AV dataset so I can reproduce which data was used to train a specific model checkpoint","I need to track which annotators labeled which scenes and maintain quality metrics across a 100k+ frame dataset"],"best_for":["autonomous vehicle companies building perception datasets","robotics teams training 3D object detection models","enterprises deploying safety-critical computer vision systems"],"limitations":["3D annotation is slower than 2D — expect 5-10 minutes per frame for complex scenes vs 1-2 minutes for 2D images","LiDAR annotation requires specialized training and expertise; annotator pool is smaller and more expensive","sensor fusion annotation (coordinating labels across camera, LiDAR, radar) adds 30-50% overhead compared to single-sensor annotation"],"requires":["multi-modal sensor data (camera images + LiDAR point clouds minimum)","calibration parameters for sensor fusion (intrinsics, extrinsics)","3D bounding box schema (class definitions, size ranges)","minimum dataset size of 1,000 frames for meaningful versioning"],"input_types":["camera images (JPEG, PNG, raw sensor format)","LiDAR point clouds (PCD, LAS, or proprietary formats)","radar data (if applicable)","sensor calibration matrices","3D scene metadata (location, weather, time of day)"],"output_types":["3D bounding box annotations (center, dimensions, rotation)","per-frame object tracking IDs (for temporal consistency)","sensor fusion labels (cross-modal associations)","dataset version manifests (JSON with frame checksums and annotation metadata)","quality metrics per annotator and per scene"],"categories":["data-processing-analysis","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"scale-ai__cap_4","uri":"capability://tool.use.integration.api.driven.annotation.workflow.orchestration","name":"api-driven annotation workflow orchestration","description":"Exposes REST and GraphQL APIs for programmatic submission of annotation tasks, status polling, and result retrieval, enabling integration into ML pipelines and CI/CD workflows. Supports batch submission with configurable callbacks, webhook notifications on task completion, and structured result formatting for direct ingestion into training pipelines without manual export/import steps.","intents":["I want to integrate annotation into my ML training pipeline so new data is automatically labeled and fed to model retraining","I need to submit annotation tasks from my data processing script and poll for results without using the web dashboard","I want to receive webhook notifications when annotation batches complete so I can trigger downstream model training"],"best_for":["ML engineers building automated data labeling pipelines","teams running continuous model retraining with fresh labeled data","enterprises integrating Scale into existing data infrastructure (Airflow, Kubernetes, etc.)"],"limitations":["API rate limits (typically 100 requests/minute for standard tier) may require batching for high-volume submissions","webhook delivery is not guaranteed — requires client-side retry logic for critical workflows","result retrieval is asynchronous; no synchronous blocking API for immediate results"],"requires":["API key with appropriate scopes (task submission, result retrieval)","HTTP client library (requests, httpx, curl, etc.)","understanding of Scale's task schema and result format","webhook endpoint (HTTPS) if using callback notifications"],"input_types":["JSON task definitions (image URLs, annotation schema, metadata)","batch submission payloads (up to 10k tasks per request)","callback URLs for webhook notifications"],"output_types":["task IDs for tracking","task status (queued, in_progress, completed, failed)","structured annotation results (JSON matching submitted schema)","metadata (annotator ID, completion time, quality scores)"],"categories":["tool-use-integration","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"scale-ai__cap_5","uri":"capability://data.processing.analysis.custom.annotation.schema.definition.and.validation","name":"custom annotation schema definition and validation","description":"Allows teams to define custom annotation schemas (hierarchical taxonomies, conditional fields, multi-type labels) through a visual builder or JSON schema format, with automatic validation to ensure annotators provide complete and consistent labels. Supports schema versioning and migration, allowing schema changes without invalidating previously labeled data.","intents":["I need to define a custom annotation schema for my domain-specific task without writing code","I want to enforce that certain fields are only required when other fields have specific values (conditional logic)","I need to update my annotation schema mid-project and migrate existing labels to the new schema"],"best_for":["teams with domain-specific annotation requirements not covered by standard templates","enterprises managing multiple annotation projects with different schemas","research teams iterating on annotation design during dataset creation"],"limitations":["complex conditional schemas (>10 conditional branches) can confuse annotators and reduce agreement","schema migration for large datasets (100k+ examples) may require manual review of edge cases","no built-in schema optimization — teams must manually test schemas to find clarity issues"],"requires":["understanding of annotation task requirements (what to label, how to label it)","JSON schema knowledge (if using JSON schema format) or access to visual builder UI"],"input_types":["visual schema builder interactions (UI-based)","JSON schema definitions (programmatic)","schema templates (pre-built for common tasks)"],"output_types":["validated annotation schema (JSON)","schema version history","annotator-facing schema documentation (auto-generated)"],"categories":["data-processing-analysis","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"scale-ai__cap_6","uri":"capability://data.processing.analysis.inter.annotator.agreement.measurement.and.conflict.resolution","name":"inter-annotator agreement measurement and conflict resolution","description":"Automatically calculates agreement metrics (Cohen's kappa, Fleiss' kappa, Krippendorff's alpha) across multiple annotators on the same examples, identifies disagreement patterns, and routes conflicting labels to expert reviewers for adjudication. Provides dashboards showing agreement trends over time and per-annotator reliability scores.","intents":["I want to measure whether my annotation task is clear enough by checking if annotators agree on the same examples","I need to identify which annotators are making systematic errors and retrain or remove them","I want to automatically escalate ambiguous examples to expert reviewers instead of accepting low-agreement labels"],"best_for":["teams building high-quality training datasets where agreement is a proxy for label quality","enterprises with regulatory requirements to document annotation consistency","research teams studying annotation task design and clarity"],"limitations":["agreement metrics assume independent annotators; if annotators discuss examples, agreement will be artificially high","some tasks (subjective tasks like rating helpfulness) naturally have lower agreement (60-70%) even with clear rubrics","expert adjudication adds 20-30% overhead to annotation timeline"],"requires":["minimum 3 annotators per example for meaningful agreement statistics","at least 50-100 examples with multiple annotations to calculate reliable metrics","expert reviewers available for conflict resolution (if using adjudication)"],"input_types":["multiple annotations per example (from different annotators)","annotation schema (to understand label types)"],"output_types":["agreement scores (Cohen's kappa, Fleiss' kappa, etc.)","per-annotator reliability scores","disagreement flags and conflict reports","trend dashboards (agreement over time)","adjudication decisions (expert-resolved labels)"],"categories":["data-processing-analysis","safety-moderation"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"scale-ai__cap_7","uri":"capability://automation.workflow.managed.workforce.scheduling.and.capacity.planning","name":"managed workforce scheduling and capacity planning","description":"Manages Scale's internal annotation workforce, automatically routing tasks to available annotators based on skill level, language, domain expertise, and current workload. Provides capacity forecasting and SLA management, allowing clients to specify turnaround time requirements (e.g., 48-hour completion) and Scale automatically allocates workforce to meet commitments.","intents":["I need to annotate 10,000 images in 2 weeks but don't know how many annotators I need or how to manage them","I want to ensure my annotation tasks are completed within a specific SLA (e.g., 48 hours) without manually coordinating with annotators","I need specialized annotators (e.g., medical imaging experts) for my domain-specific task"],"best_for":["enterprises without in-house annotation teams who need predictable turnaround times","teams with variable annotation volume that would be inefficient to staff internally","projects requiring specialized domain expertise (medical, legal, autonomous vehicles)"],"limitations":["SLA guarantees come at a premium cost (typically 20-40% markup over standard pricing)","specialized annotators (e.g., medical experts) have limited availability and may require 2-4 week lead time","workforce capacity fluctuates with demand across all Scale clients; peak periods may have longer wait times"],"requires":["clear task definition and annotation schema (so Scale can estimate complexity and allocate resources)","realistic turnaround time expectations (minimum 24-48 hours for most tasks)"],"input_types":["annotation tasks with metadata (complexity, domain, language requirements)","SLA requirements (desired completion time)"],"output_types":["task assignment confirmations","real-time progress tracking (% complete)","completion notifications with SLA compliance status"],"categories":["automation-workflow","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"scale-ai__cap_8","uri":"capability://safety.moderation.data.security.and.compliance.certification.management","name":"data security and compliance certification management","description":"Provides SOC 2 Type II, FedRAMP, HIPAA, and GDPR compliance certifications with encrypted data handling, secure data deletion, and audit logging. Manages data residency requirements (e.g., data must stay in US regions) and provides detailed audit trails showing which annotators accessed which data and when.","intents":["I need to label sensitive healthcare data but require HIPAA compliance and encrypted data handling","I'm a government contractor and need FedRAMP-certified annotation services","I need to prove to auditors that my annotation data was handled securely and deleted after use"],"best_for":["healthcare and biotech companies handling PHI (Protected Health Information)","government agencies and contractors with FedRAMP requirements","enterprises in regulated industries (finance, legal) with strict data governance"],"limitations":["compliance certifications add 15-30% cost premium over standard annotation pricing","FedRAMP compliance requires government approval and may take 3-6 months for new clients","data residency restrictions (e.g., US-only) limit annotator pool and may increase turnaround times"],"requires":["compliance requirements clearly defined upfront (HIPAA, FedRAMP, GDPR, etc.)","data classification (what data is sensitive and requires special handling)","audit and compliance team to review Scale's certifications and controls"],"input_types":["sensitive data (healthcare, government, financial)","compliance requirement specifications"],"output_types":["audit logs (who accessed what data, when)","compliance attestations (SOC 2, FedRAMP, HIPAA)","data deletion confirmations","encryption certificates and key management records"],"categories":["safety-moderation","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"scale-ai__cap_9","uri":"capability://data.processing.analysis.active.learning.task.prioritization.and.uncertainty.sampling","name":"active learning task prioritization and uncertainty sampling","description":"Integrates with client ML models to identify which unlabeled examples would be most valuable to label next, using uncertainty sampling and model-based prioritization. Automatically submits high-value examples for annotation and tracks how much each labeled example improves model performance, enabling data-efficient labeling strategies.","intents":["I want to label only the most informative examples instead of randomly sampling from my dataset to reduce labeling costs","I want to measure how much each labeled example improves my model's performance so I can optimize labeling ROI","I want to automatically identify edge cases and hard examples that my model struggles with and prioritize them for labeling"],"best_for":["ML teams with large unlabeled datasets who want to minimize labeling costs","startups with limited labeling budgets who need to maximize data efficiency","research teams studying active learning strategies"],"limitations":["requires integration with client's ML model and training pipeline; not a standalone feature","uncertainty sampling works best for classification tasks; less effective for structured prediction (NER, object detection)","model performance improvements from active learning are typically 10-20% better than random sampling, not orders of magnitude"],"requires":["trained ML model that can produce uncertainty estimates (confidence scores, entropy, etc.)","unlabeled dataset with at least 1,000 examples","ability to retrain model after each labeling batch (for feedback loop)"],"input_types":["unlabeled examples","model predictions and confidence scores","model architecture and weights (for uncertainty estimation)"],"output_types":["ranked list of examples by estimated value","performance improvement metrics (before/after labeling)","active learning curves (cost vs accuracy)"],"categories":["data-processing-analysis","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"scale-ai__headline","uri":"capability://data.processing.analysis.enterprise.data.labeling.and.ai.infrastructure.platform","name":"enterprise data labeling and ai infrastructure platform","description":"Scale AI is an enterprise-grade platform focused on providing high-quality data labeling and annotation services for AI applications, including computer vision, NLP, and generative AI, ensuring a human-in-the-loop approach for optimal model training.","intents":["best data labeling platform","data annotation for AI model training","enterprise data labeling solutions","AI infrastructure for computer vision","human-in-the-loop annotation services"],"best_for":["enterprise applications","government projects"],"limitations":[],"requires":[],"input_types":[],"output_types":[],"categories":["data-processing-analysis"],"confidence":0.5,"matches":0,"success_rate":0}],"trust":{"score":56,"verified":false,"data_access_risk":"low","permissions":["image dataset in JPEG, PNG, or WebP format","annotation schema defined in Scale's JSON schema format or via web UI","API key for programmatic access (if using API rather than web dashboard)","text data in plain text, CSV, or JSON format","entity schema or classification taxonomy defined upfront","minimum batch size of 100 examples for efficient processing","clear specification of required language and dialect","annotation schema adapted for language-specific considerations (e.g., grammatical structures, cultural context)","minimum batch size of 100 examples per language for efficient processing","trained ML model that can produce predictions on unlabeled data"],"failure_modes":["consensus-based QA adds 20-40% latency to annotation cycles compared to single-pass labeling","custom annotation schemas require JSON schema definition and may need 1-2 iteration cycles to optimize for workforce understanding","no real-time streaming annotation — batches must be submitted and processed asynchronously","hierarchical entity schemas with >50 entity types may cause annotator confusion and require extensive training","no built-in active learning — cannot automatically select most uncertain examples for labeling","turnaround time for large batches (10k+ examples) is 3-7 days depending on complexity and workforce availability","rare languages (< 1 million speakers) may have limited annotator availability and higher costs","dialect-specific annotation (e.g., Moroccan Arabic) requires specialized annotators and may have 2-4 week lead times","quality assurance for language-specific tasks is harder to automate; requires native speaker expert review","pre-labeling quality depends on model accuracy; if model is <70% accurate, pre-labels may introduce more errors than they save","builder identity is not verified yet","no observed match outcomes yet"],"rank_breakdown":{"adoption":0.7,"quality":0.9,"ecosystem":0.15000000000000002,"match_graph":0.25,"freshness":0.75,"weights":{"adoption":0.3,"quality":0.25,"ecosystem":0.15,"match_graph":0.25,"freshness":0.05}},"observed_outcomes":{"matches":0,"success_rate":0,"avg_confidence":0,"top_intents":[],"last_matched_at":null},"maintenance":{"status":"active","updated_at":"2026-05-24T12:16:25.061Z","last_scraped_at":null,"last_commit":null},"community":{"stars":null,"forks":null,"weekly_downloads":null,"model_downloads":null,"model_likes":null}},"distribution":{"claim_url":"https://unfragile.ai/submit?claim=scale-ai","compare_url":"https://unfragile.ai/compare?artifact=scale-ai"}},"signature":"wxgDoiNO1wt2oUbblJLYyjKYzhpjo83fbWZion2hzq+epnPxXrLUujQwow4h95zb2E5iqDW3Z4qidYf+/7qYCA==","signedAt":"2026-06-22T17:43:50.737Z","signedBy":"unfragile.ai","version":1},"_links":{"self":"https://unfragile.ai/api/v1/passport/scale-ai","artifact":"https://unfragile.ai/scale-ai","verify":"https://unfragile.ai/api/v1/verify?slug=scale-ai","publicKey":"https://unfragile.ai/api/v1/trust-passport-public-key","spec":"https://unfragile.ai/trust","schema":"https://unfragile.ai/schema.json","docs":"https://unfragile.ai/docs"}}