Modern AI agents are no longer bound by generic LLM capabilities.
With Amazon’s Nova models integrated into Lyzr Agent Studio, enterprises can now build agents that fit like a glove, optimized for specific tasks, deployed securely within AWS, and tuned to balance speed, accuracy, and cost.
This guide walks through how to build and deploy custom AI agents using Nova models via Lyzr, with a detailed look at the model selection, practical use cases, and enterprise-grade deployment.
Nova Foundation Models: Overview and Launch
Announced at AWS re:Invent 2024, the Nova family represents Amazon’s in-house foundational models designed specifically for enterprise-grade inference within the AWS ecosystem.
These models address both core content understanding and generative tasks, offering native AWS integration, performance tuning, and fine control over data locality and governance.
Nova Model Families
| Model | Input Modalities | Output | Key Usage |
| Nova Canvas | Text, Image prompt | Image, Video | Creative asset generation, visual design |
| Nova Reel | Text, Video prompt | Video | Synthetic video generation, media creation |
Why Lyzr AI Agents for Enterprises?
Lyzr’s enterprise-grade architecture is built for efficient scaling and high performance, ensuring AI agents run reliably even under demanding workloads.
- Effortless scaling: Expand your AI infrastructure smoothly to meet growing business demands.
- Optimized performance: Compatible with 100+ LLMs and 20+ vector databases to deliver fast, efficient execution.
- Custom integrations: Connect with major cloud platforms and enterprise systems for fast and flexible deployment.
Lyzr’s enterprise-ready architecture is designed for scalable performance, allowing AI agents to run efficiently, even during high-demand operations.
- Centralized intelligence: Continuously tracks agent performance to optimize outcomes across the board.
- Predictive analysis: Analyzes patterns in real time to forecast issues and recommend the best course of action.
- Autonomous decision-making: OGI operates independently to implement strategies, driving optimal outcomes for the organization.
Why Nova Models for Enterprise AI Agents?
Amazon’s Nova models are first-party foundational models developed in-house by AWS, not external vendors. This gives them three critical advantages:
- Fully hosted on AWS infrastructure: All inference is processed inside the AWS cloud, eliminating the need for third-party routing.
- Billing via native AWS services: Nova models are billed as native services, not marketplace add-ons. This qualifies for AWS Enterprise Discount Programs (EDP) and centralized billing.
- Integrated with AWS Guardrails and IAM: Nova models can be secured, governed, and audited using native AWS services like IAM, KMS, VPC, and Bedrock Guardrails.
Nova vs OpenAI vs Claude: Model Comparison
When deploying AI agents in production environments, enterprises need more than just model performance. Cost predictability, latency, scalability, and compliance readiness are critical to operational success. Here’s how Nova, OpenAI, and Claude stack up across key enterprise priorities.
1. Cost Efficiency
Enterprises deploying agents at scale need models that are not only effective but economically viable. Nova delivers high performance without premium pricing, thanks to its native AWS deployment, avoiding third-party markups and API costs.
| Aspect | Nova (via Lyzr on AWS) | OpenAI (GPT models) | Claude (Anthropic) |
| Inference Cost | Lower cost due to AWS-native deployment | Higher due to closed API model and markup | Premium pricing tied to proprietary APIs |
| Billing | Integrated with AWS billing and EDP drawdowns | Separate contract and invoicing | Separate vendor billing |
| Scalability ROI | Economically scales with enterprise usage | Escalating costs with increased token usage | Expensive for sustained or high-volume usage |
2. Latency & Performance
Speed is critical for agents embedded in live workflows. Nova delivers faster response times by running directly within the enterprise’s AWS environment, avoiding the latency added by external API calls.
| Aspect | Nova (via Lyzr on AWS) | OpenAI (GPT models) | Claude (Anthropic) |
| Execution Time | Sub-second, inside enterprise VPC | Slower due to external API routing | Similar external latency, sometimes higher |
| Throttling Risks | No external rate limits | Subject to OpenAI rate limits | Throttled on free/standard tiers |
| Consistency | High, with direct control | May vary due to shared infrastructure | Less predictable under heavy load |
3. Scalability & Control
Nova integrates directly with AWS-native services like Auto Scaling, CloudWatch, and Lambda, giving enterprises total control over model scaling and resource allocation. With OpenAI and Claude, scaling is gated by vendor limits and pricing tiers.
| Aspect | Nova (via Lyzr on AWS) | OpenAI (GPT models) | Claude (Anthropic) |
| Scaling Method | Fully integrated with AWS auto-scaling | Controlled by OpenAI service limits | Manual quota upgrades required |
| Observability | Uses native AWS monitoring & logs | Limited visibility through vendor dashboards | Minimal enterprise-grade monitoring tools |
| Regional Availability | Configurable by enterprise inside AWS regions | Dependent on OpenAI’s hosted infrastructure | Fixed endpoints with regional constraints |
4. Compliance & Security
For enterprises in regulated industries, data governance is non-negotiable. Nova never moves data outside AWS, ensuring compliance with internal and industry-wide policies. OpenAI and Claude require data to exit the enterprise perimeter, raising legal and compliance overhead.
| Aspect | Nova (via Lyzr on AWS) | OpenAI (GPT models) | Claude (Anthropic) |
| Data Residency | Data stays within enterprise-owned AWS VPC | Data exits to OpenAI infrastructure | External API endpoints |
| Encryption | At rest and in transit, with AWS-native tooling | Vendor-managed, limited transparency | Vendor-controlled encryption |
| Compliance Readiness | Covered by AWS terms and existing legal approvals | Requires new vendor risk assessment | Requires separate compliance workflows |
| IAM & Access Control | Governed by AWS IAM and KMS | Vendor-specific access management | External access control, less granular |
5. Deployment & Vendor Integration
Nova minimizes procurement complexity. Since it’s fully native to AWS, there’s no need for separate legal contracts or vendor onboarding, accelerating adoption without red tape.
| Aspect | Nova (via Lyzr on AWS) | OpenAI (GPT models) | Claude (Anthropic) |
| Onboarding Effort | Zero—covered under existing AWS relationship | Requires vendor review and contract | Separate vendor onboarding required |
| Legal & Procurement | Simplified via AWS terms and conditions | New contract required | Independent legal and security evaluation |
| Support Structure | AWS enterprise support + Lyzr platform support | OpenAI-specific support | Anthropic-managed support stack |
Lyzr + Nova = Low-Code AI Agents Built for Production
Lyzr Agent Studio abstracts agent development through two main interfaces:
- No-Code Agent Builder for business users to create and deploy agents using prebuilt logic and model choices.
- Developer APIs for teams to programmatically construct, monitor, and extend agents.
Lyzr integrates directly with Amazon Bedrock, enabling developers to select from available Nova models. All agent logic, inference, and orchestration run within AWS, either in Lyzr’s isolated AWS account or in the customer’s private VPC.
Currently Supported Nova Models in Lyzr
Step-by-Step: Building an AI Agent Using Lyzr + Nova
1. Define Your Use Case
The first step is identifying the role of your agent. Examples include:
- A customer service bot that handles queries and follows up
- An internal KYC document summarizer
- A CRM-connected sales assistant
Once the use case is locked, model selection becomes easier.
2. Choose the Right Nova Model
Refer to the table below to map your agent to the appropriate Nova model:
| Use Case | Recommended Model | Why? |
| Notification agents | Nova Micro | High throughput, low latency |
| KYC Document Parser | Nova Lite | Multimodal input with good reasoning |
| CRM-integrated assistant | Nova Pro | Handles workflows, API orchestration |
3. Use Lyzr Agent Studio to Build the Agent
Lyzr supports both:
- No-Code Builders: Drag-and-drop interface for building actions, triggers, and workflows.
- APIs: For embedding agents directly into existing software.
Within the studio:
1. Select your Nova model.
2. Select core features
3.. Add tool integrations (Slack, Github, etc.).
100% AWS-Native Architecture
Enterprises building agents with Nova via Lyzr benefit from a fully AWS-native architecture that aligns with existing security, compliance, and procurement standards, without introducing any new external dependencies.
Data Stays Within AWS: All agent execution happens entirely within the customer’s AWS Virtual Private Cloud (VPC). There are no external API calls or third-party data transfers, ensuring that sensitive enterprise data never leaves the AWS environment. This architecture minimizes risk while meeting the strictest internal data governance requirements.
Unified Billing and Procurement: Nova is available as a native AWS service. That means no additional contracts or vendor onboarding. Enterprises can leverage their existing Enterprise Discount Program (EDP) with AWS for Nova usage, allowing all costs to roll into their consolidated AWS billing. Procurement becomes faster, cleaner, and easier to manage.
Built-In Guardrails for Responsible AI: Security and governance are integrated into every layer. Agent actions are governed by AWS Identity and Access Management (IAM) policies, encrypted using AWS Key Management Service (KMS), and logged using native audit tools. Responsible AI tooling is also included to enforce enterprise-grade control and traceability across all agent activity.
Simplified Legal and Compliance: Because Nova is an AWS-native offering, it falls under AWS’s existing terms and conditions. This eliminates the need for separate legal review or compliance checks. Enterprises can deploy agents with confidence, knowing they’re fully covered under their current AWS agreements—no new vendors, no extra paperwork.
Final Thoughts
With Nova models available in Lyzr, developers and enterprises can now:
- Launch intelligent agents in hours, not weeks
- Match task complexity to the right model (Micro, Lite, Pro)
- Stay compliant with AWS-native execution
As Nova Sonic and future models join the lineup, expect even richer voice agents, creative copilots, and full-stack orchestration possibilities. But today, even with just Micro, Lite, and Pro, teams can build fast, secure, production-ready AI agents with zero infrastructure friction.
Ready to start building? Head to Lyzr Agent Studio and choose your Nova model.
Book A Demo: Click Here
Join our Slack: Click Here
Link to our GitHub: Click Here
