Product: ATLAS
Publisher: VedaCLOUD.AI
Release document type: Platform capability overview
Version: 1.0
Release date: July, 22 2026

1. Overview#

VedaCLOUD ATLAS is an enterprise AI platform designed to help organizations transform enterprise information into searchable knowledge, governed AI workflows, and actionable intelligence.

The platform brings together document intelligence, retrieval-augmented generation (RAG), hybrid search, knowledge graphs, model orchestration, agentic workflows, identity and access controls, and operational tooling within a modular, containerized architecture.

This release overview summarizes the platform capabilities documented in the current ATLAS reference implementation. Feature availability may depend on deployment configuration, connected services, model providers, and the maturity of individual integrations.

2. Highlights#

Enterprise knowledge and document intelligence#

• Document ingestion and processing for common enterprise formats, including PDF, Word, Excel, PowerPoint, HTML, text, Markdown, CSV, images, and supported email formats.
• Text extraction with OCR, layout-aware processing, reading-order handling, and table-oriented extraction capabilities.
• Document chunking, metadata processing, embedding generation, and indexing workflows.
• Support for English, Hindi, and additional Indian scripts through configured OCR capabilities.

Retrieval-augmented generation and hybrid search#

• Retrieval workflows combining semantic and keyword-based search.
• Context retrieval to support grounded responses from enterprise knowledge sources.
• Source references that can include document, page, and paragraph information where available.
• Permission-aware retrieval designed to respect applicable access controls.
• Optional web research through configured search services, subject to deployment and network policy.

Knowledge graph capabilities#

• Entity and relationship extraction from processed content.
• Provenance-aware knowledge representation.
• Graph exploration and multi-hop relationship traversal.
• Integration between relational records and graph-based knowledge operations, with documented fallback behavior when the graph service is unavailable.

Model orchestration#

• Integration options for local model-serving environments, including Ollama, vLLM, and LM Studio.
• Configurable integrations with supported external model providers, including OpenAI, Azure OpenAI, Google, Anthropic, and Hugging Face.
• Model-routing policies that can consider configured priorities, latency, cost, quality, load, and failover behavior.
• Configurable model-usage limits.

External provider availability, credentials, routing behavior, and data-processing boundaries depend on the selected configuration.

Agent Studio and workflow automation#

• Natural-language-assisted workflow creation and a visual workflow editor.
• Workflow steps for model interaction, knowledge retrieval, graph queries, configured web research, HTTP APIs, MCP tools, and sandboxed Python execution.
• Support for conditional logic, variables, templates, structured outputs, and child-agent workflows.
• Workflow version history, rollback, and step-level debugging capabilities.
• Configurable workflow access scopes and execution controls.

Execution limits, available tools, and isolation behavior depend on the deployment and workflow configuration.

3. Enterprise AI capabilities#

Document AI#

ATLAS supports configurable document-processing workflows for classification, information extraction, validation, human review, document comparison, and structured reporting.

The results depend on document quality, extraction configuration, model selection, and validation rules.

Identity and access management#

Documented platform capabilities include authentication, role-based access control (RBAC), configurable multi-factor authentication policies, and options for single sign-on and directory synchronization.

Identity-provider integrations and policies must be configured and tested for the target environment.

Privacy and AI safety controls#

The platform includes configurable controls intended to support safer enterprise AI operation, including:

• Protection of provider credentials and support for a secrets-management integration.
• PII masking for applicable prompt-processing paths.
• Prompt-injection, jailbreak, and output-safety controls.
• Configurable tool access and workflow execution restrictions.
• Answer-verification capabilities and permission-aware knowledge retrieval.

These controls reduce specific risks but do not guarantee the prevention of every security, privacy, or model-related failure. Their effectiveness must be validated against the intended use cases.

Evaluation and quality assurance#

ATLAS includes capabilities for working with evaluation datasets, assessing retrieval quality, checking grounding, running regression tests, and comparing model behavior.

Evaluation results depend on dataset coverage, test design, configuration, and the quality criteria selected.

4. Deployment and infrastructure#

The documented reference architecture uses Docker Compose to orchestrate modular application services and supporting infrastructure.

The technology foundation includes:

Component	Role
Go, Python, and TypeScript services	Application and platform services
PostgreSQL	Relational records and core system data
Redis	Caching, sessions, rate limiting, and queue-related functions
MinIO	Object and file storage
Weaviate	Vector and keyword-oriented retrieval
Memgraph	Graph-based knowledge operations
SearXNG	Optional private metasearch integration

The reference architecture supports CPU-based deployment, with optional NVIDIA GPU acceleration for suitable workloads. Local-model and hybrid local/external-provider configurations are available subject to setup.

An air-gapped deployment mode is documented, with external web research and outbound model-provider access disabled in that mode. Individual deployment profiles and network rules must be reviewed to confirm the intended isolation boundary.

Important: The reference architecture is not, by itself, evidence of a production certification, a specific service-level commitment, or an independently verified security posture.

5. Operations and administration#

Documented operational capabilities include:

• Service health and monitoring functions.
• Metrics and gateway audit logging.
• Workflow execution history.
• Backup and restore procedures for core records, uploaded files, and configuration.
• Reindexing and knowledge-data recovery workflows.
• Troubleshooting and deployment guidance.

The documented recovery approach supports rebuilding vector indexes from source data and resynchronizing graph data from the system of record where applicable.

Backup frequency, retention, recovery objectives, regional recovery arrangements, monitoring coverage, and incident-response procedures must be established and validated for each production deployment.

6. Integration and interoperability#

ATLAS is organized as a modular platform with APIs, connectors, model-provider integrations, and configurable workflow tools.

Available integration paths depend on configured credentials, endpoints, network access, permissions, and connector implementation status. Not every planned connector should be assumed to be production-ready or enabled by default.

Before deployment, validate each required integration end to end, including authentication, authorization, error handling, rate limits, logging, and recovery behavior.

7. Security and deployment considerations#

Before using ATLAS for sensitive or business-critical workloads, deployment teams should review and validate:

• Identity-provider configuration, RBAC policies, and privileged access.
• Encryption in transit and at rest across all relevant services and storage layers.
• Secrets management and credential rotation.
• Data residency for application data, prompts, retrieved context, embeddings, indexes, logs, backups, and model inference.
• External model-provider data handling and contractual settings.
• Network exposure, reverse-proxy configuration, service isolation, and administrative endpoints.
• Sandbox and tool-execution restrictions.
• Audit coverage, monitoring, incident response, backup restoration, and disaster recovery.
• Resource sizing, model performance, concurrency, and operational limits.

Data residency and processing location depend on the actual infrastructure, storage configuration, network paths, and selected model providers. These properties must be verified for the target deployment rather than inferred from the platform architecture alone.

8. Known limitations and validation requirements#

The following items require deployment-specific confirmation and should not be assumed solely from the reference implementation:

1. Version-specific changes: A versioned changelog with build identifiers, release dates, and verified changes has not been supplied for this document.
2. Connector readiness: Individual integrations may require additional configuration, testing, or implementation work.
3. Production hardening: Container orchestration and application configuration require environment-specific security review before production use.
4. Model and provider behavior: Performance, routing, availability, cost, and data handling depend on selected models and provider configurations.
5. Security-control effectiveness: Guardrails, PII masking, access controls, and injection defenses require testing against the relevant threat model.
6. Availability and recovery: No specific uptime SLA, RTO/RPO, or independently verified disaster-recovery commitment is established by this overview.
7. Certifications and assurance: No external certification or independent audit is asserted by this document.

9. Recommended deployment acceptance checks#

Before promoting a particular ATLAS build into production, the deployment team should record the exact build or commit identifier and complete acceptance testing covering:

• Core service startup, health checks, and dependency availability.
• Document ingestion, OCR, indexing, retrieval, and source attribution.
• User and role permissions across documents, search, and agent workflows.
• Local and external model connectivity, including failure and timeout behavior.
• Workflow execution limits, tool permissions, and sandbox isolation.
• Logging, monitoring, audit events, and administrative access.
• Backup restoration, index reconstruction, and graph-data recovery.
• Data-flow, residency, privacy, and network-boundary verification.
• Performance, concurrency, resource consumption, and operational readiness.

Acceptance results should be recorded against the specific deployment configuration and release identifier.

10. Summary#

VedaCLOUD ATLAS provides a modular foundation for enterprise knowledge management, retrieval-grounded AI, knowledge graphs, model orchestration, and governed workflow automation. Its architecture combines configurable application services with relational, object, vector, graph, and caching infrastructure, alongside documented identity, evaluation, and operational capabilities.

The platform is designed to support a range of deployment patterns, from local-model environments to configured hybrid deployments. Production suitability depends on the selected configuration, completed integrations, security validation, and operational acceptance testing.

VedaCLOUD ATLAS — Enterprise knowledge. Governed AI. Intelligent automation.

For platform information, contact connect@vedacloud.ai.

---

Document note: This is a platform capability overview, not a verified change log for a specific software version. Publish it as a dated release only after confirming the applicable build, feature availability, and release-specific changes.