Core platform components
Encord consists of four tightly integrated product areas:| Component | Purpose |
|---|---|
| Index | Data ingestion, organization, search, and curation |
| Annotate | Annotation, review, QA workflows, and human feedback |
| Active | Model evaluation, analytics, and active learning |
| Agents | Automation, AI-assisted labeling, and model integrations |
Data architecture
How data is stored
Encord operates on a bring-your-own-storage model. Your files remain in your cloud storage (AWS S3, GCP Cloud Storage, or Azure Blob Storage) at all times. Encord registers references to those files, not copies of them. This means:- Raw data never leaves your storage environment
- You retain full ownership and control of your files
- Encord only stores metadata, labels, annotations, and embeddings
Data registration models
Standard integration (client cloud storage) Your data stays in your private cloud bucket. Encord is granted read access via a service account or IAM role. Files are streamed to the platform for display and annotation but never persisted outside your environment. Strict client external integration For highly regulated environments, Encord can operate in a mode where all data access is mediated through your own infrastructure. Data is never cached or proxied through Encord servers. Encord-managed storage Files are uploaded to and stored within Encord’s GCP environment. Suitable for teams that do not have existing cloud storage or who need the simplest possible setup. See Data Integrations for setup instructions.Deployment models
Encord offers three deployment options to match your security and infrastructure requirements.Cloud (Encord-managed)
The default deployment. Encord manages all infrastructure on Google Cloud Platform. Your data stays in your own storage bucket; Encord only accesses it to serve annotations.- Fastest to deploy
- No infrastructure management required
- SOC 2 Type II, HIPAA, GDPR compliant
- Regular updates and patches managed by Encord
VPC (Virtual Private Cloud)
Encord is deployed within your own cloud environment (AWS, GCP, or Azure). All compute and application servers run inside your network boundary.- All traffic stays within your network
- You control firewall rules, logging, and auditing
- Requires coordination with Encord for initial setup
On-premise / air-gapped
For environments with strict data residency or security requirements — including classified, regulated, or offline environments — Encord can be deployed with no outbound internet access.- No data leaves your internal network
- Suitable for defense, healthcare, and regulated industries
- Requires dedicated support engagement
Supported data types
Encord supports the following data types across Index, Annotate, and Active:| Modality | Formats |
|---|---|
| Images | jpg, png, tiff |
| Video | mp4, mov, webm, matroska, 3gp, m4a, mj2 |
| Audio | mp3, wav, m4a, flac, mp4, eac3 |
| DICOM | Standard DICOM series and studies |
| Documents | PDF, HTML, plain text |
| Point clouds | LAS, PCD, and related 3D formats |
Scalability
Encord’s architecture is designed to scale horizontally. As data volumes and team sizes grow:- Compute resources scale automatically with workload
- Annotation queues support millions of tasks across distributed teams
- Embedding and analytics pipelines process large datasets asynchronously
- API rate limits and throughput can be raised for high-volume enterprise customers
Where to go next
- Security and Compliance — certifications, access controls, and data governance
- Scaling and Operations — workforce structure, QA, and project management at scale
- Data Integrations — connecting AWS, GCP, and Azure storage
- Supported Data — full list of supported file formats