# Platform Look under the hood of the self-driving cloud.™ Learn how Sedai safely reduces cloud costs, improves app performance, & increases availability. ## Welcome Inside Your Future Cloud Zero wasted costs. Zero IaC drift. Zero performance risk. Sedai drives your cloud like a supercar — based on your directions. ## Autonomous Optimization Engine Sedai works by learning how your apps actually behave: traffic patterns, dependencies, golden signals, & more. That's the key to reducing costs, without risking performance. Our Autonomous Optimization Engine uses this app intelligence to make gradual changes, while performing continuous safety checks. That's how Sedai optimizes every layer of the cloud — from microservices to major purchasing decisions. ## Enterprise-Grade Controls for the Cloud Sedai gives you complete control over your cloud infrastructure. All you have to do is set SLOs, create guardrails, & choose Sedai's driving mode. Our AI makes the right optimizations to achieve your goals — based on the real-time behavior of your apps. ## A Trusted Path to Autonomy Sedai is designed to meet teams where they are and let you adopt autonomy at your own pace. You can run different modes across services, environments, or teams, and change modes at any time. You're always in control. ### Datapilot Sedai observes your environment and builds understanding. We surface cost and performance insights and explain exactly what actions we would take, without making any changes. ### Copilot Sedai recommends actions, but now you can approve and apply them with a click. You stay in control, while we remove the manual effort and accelerate execution. ### Autopilot Sedai safely takes action on its own, within the policies and guardrails you define. Optimization becomes continuous and real-time, with every change explainable and fully reversible. ## Performance Results Our customers set new benchmarks for efficiency, latency, & availability. Ready to see your true peak performance? LOWER CLOUD COSTS LESS MANUAL OPTIMIZATION WORK INCIDENTS ACROSS 25M+ AUTONOMOUS ACTIONS ## Optimize Your Whole Stack ### Kubernetes Sedai rightsizes workloads, tunes scaling, and optimizes nodes and clusters to cut costs while meeting SLOs across managed (EKS, AKS, GKE) and self-managed/on-prem Kubernetes. Rightsize requests/limits from real utilization and latency; apply vertical tuning (CPU/mem) and in-place pod resizing where supported; verify changes against SLOs/error budgets. Improve bin-packing while honoring affinities/taints/PDBs; remove idle or underused nodes and right-size node groups/instance mixes — all within customer-defined guardrails that maintain resilience. Tune HPA/VPA targets, min/max, step sizes, and cooldowns; schedule policies by time-of-day/seasonality and adjust around releases; prevent thrash and auto-rollback if SLOs drift. ### Virtual Machines Sedai rightsizes instances & tunes autoscaling policies to balance performance, resilience, & cloud spend across environments — without manual analysis or tuning. Recommend the optimal instance family & size (including newer-generation or burstable types) based on CPU, memory, & I/O profiles. Plus, schedule changes during maintenance windows and validate them against latency & SLO impact. Calibrate VM utilization, update scaling policies, & adjust instance mixes while honoring user-provided guardrails to maintain resilience & performance. Automatically start & stop non-production environments on calendars, detect idle or underutilized VMs, and surface cleanup opportunities to reduce waste & improve efficiency. ### Data Platforms Sedai cuts data platform compute costs by right-sizing clusters, tuning autoscaling, & aligning configurations with real workload behavior. Teams get lower cost per run & clear visibility into realized savings. Analyze long-running interactive clusters to right-size configurations, tune autoscaling bounds, & reduce unnecessary runtime. Eliminate overprovisioning & wasted cluster hours caused by user-driven cluster sprawl & inefficient defaults. Evaluate historical job runs to right-size driver and executor cores and memory, tune autoscaling bounds, and select appropriate node types. Reduce cost per run by removing excess capacity and matching compute to real workload behavior. Attribute Databricks compute costs at the cluster and job level, including DBUs and underlying cloud spend. Apply provider discounts and historical billing data to show true realized cost and provide accurate before-and-after savings measurement. ### GPU Sedai optimizes GPU infrastructure design and workload placement to maximize utilization, reduce cloud spend, and ensure high-performance availability for AI workloads. Identify and execute node-level optimization opportunities by analyzing total vs. idle devices; visualize GPU and memory capacity alongside cost savings; manage evictable or cordoned nodes to support efficient cluster compaction. Access granular configuration details in the node side drawer, including instance types, GPU models, device count, and memory; monitor key utilization metrics such as GR Engine and SM Utilization alongside labels and taints. Right-size GPU workloads with targeted Multi-Instance GPU (MIG) recommendations (e.g., converting full-GPU limits to MIG profiles like mig-1g.5gb); view estimated monthly savings and impact analysis directly alongside other resource recommendations. ### LLM & Agents Sedai acts as a smart routing and optimization layer for AI agents, managing model selection and lifecycle behavior to balance performance, cost, and accuracy across diverse providers. Automatically select the best-suited models for an agent’s domain and actions—without requiring prior evaluation datasets. Dynamically prioritize accuracy, latency, or cost, and support cross-provider routing with adaptive model selection. Improve agent resilience with group-wise load balancing, automatic fallbacks, and intelligent retries. Natively support text streaming, tool calling, and structured reasoning calls to deliver seamless user experiences. Manage agents within dedicated projects with granular cost and usage visibility. Enforce compliance through org-level access controls (allow/deny models) and centralize provider key management for secure, consistent operations. ### Serverless Sedai tunes memory, concurrency, and scheduling to hit latency targets at the best price-per-ms across AWS Lambda, Azure Functions, and Cloud Functions. Adjust memory to match CPU/I/O profile and minimize duration cost; right-size timeouts to prevent waste; verify impact on p95 latency and error budgets. Set (and schedule) provisioned/reserved concurrency for peak windows, pre-warm critical paths, and cap bursts to protect downstream services. Apply backoff, jitter, and DLQs per source (e.g., SQS/EventBridge, Pub/Sub); enforce idempotency to stop hot loops; auto-pause non-prod during off-hours. ### Storage Sedai trims storage spend while meeting retention and performance requirements—automating lifecycle tiering, right-sizing IOPS/throughput, and cleaning up waste across clouds. Analyze bucket usage patterns and data access frequency to apply optimal lifecycle policies, shifting objects into infrequent-access or archival tiers. Recommend and leverage cloud-native features (e.g., Intelligent Tiering, Auto-Class) to reduce cost automatically. Right-size volume type and provisioned IOPS/throughput to match workload demands. Reclaim waste by detecting and removing unattached or underused volumes. Automatically create protective snapshots before applying right-sizing or cleanup actions. Manage snapshot lifecycles by expiring obsolete backups & retaining only what’s required for safety and compliance. Every optimization is reversible, reducing operational risk while improving cost efficiency. ### Databases Sedai tunes database capacity, concurrency, and storage to hit latency/throughput targets at lower cost, starting with managed engines like Amazon RDS. Recommend the right instance class/size and adjust I/O capacity to maintain healthy utilization & tail latency, without over-provisioning. Verify impact against SLO/error budgets. Select appropriate storage class & provisioned IOPS/throughput, set autoscaling thresholds, and optimize backup/retention policies. Remove stale/orphaned snapshots to cut waste. Calibrate connection pool sizes, max connections, & throttles/backoff to prevent saturation and queuing — coordinating limits with application tiers where needed. ## Sedai is the Safe Way (Because we literally patented safety.) Yes, there are other solutions for cloud optimization. No, they’re not safe. Sedai has eight U.S. patents on the ability to take autonomous action, without ever causing an incident in production. Zero. That’s a promise only we can make. ## Ready to Start? ## FAQ about Sedai ### How much does Sedai cost? How does pricing work? Sedai uses a simple, volume-based pricing model, designed to ensure our customers achieve a positive ROI. For example, we charge 1 Sedai Billing Unit (SBU) for each vCPU that our platform manages. There are no overage penalties — if you exceed your monthly commitment, you simply pay the same rate for the extra usage. You can also adjust your commitment, as your needs change over time. Please [book a demo](http://sedai.io/demo) to discuss our specific pricing options & discounts. ### How does Sedai make sure that optimizations never cause availability issues or degrade performance? Safe optimization is our secret sauce. When Sedai identifies a potential cost or performance improvement, our platform makes that optimization in gradual steps, while continuously validating the impact of the change on your environment. For example, if Sedai determines it can lower the memory limit of a container without impacting performance, it won’t make a big change all at once, from say 12 to 4 GB. Instead, Sedai will gradually lower the memory to achieve maximum cost savings, until the exact point where performance is impacted. ### Which cloud providers does Sedai support? And what specific services can Sedai optimize? Sedai supports the widest range of cloud services & on-prem workloads in our industry. This includes all the major cloud providers: AWS, Microsoft Azure, Google Cloud Platform, & Oracle Cloud Infrastructure. We’re the experts at optimizing Kubernetes — think EKS, AKS, & GKE, as well as on-prem, self-managed clusters. But our platform is unique in going way beyond K8s. Sedai also optimizes storage, data platforms, and AI/ML workloads. Check out our [integrations page](http://sedai.io/integrations) for the full list. ### What makes Sedai different from competitors? Why choose Sedai? Only Sedai deeply understands the behavior of your applications — which is why our platform never causes performance or availability issues. Other cloud optimizers make risky changes, since they only look at surface-level metrics like CPU & memory utilization. Meanwhile, Sedai uses patented ML models to learn what your apps actually need, by studying traffic patterns, dependencies, & golden signals. This approach makes Sedai — and only Sedai — safe in production. Our platform has performed more than 25 million optimizations, with 0 incidents. ### Will I be able to control Sedai’s actions? What if we want to roll back a change? Sedai offers three operating modes that give you full control. In Datapilot mode, Sedai provides read-only recommendations, for you to implement. In Copilot mode, Sedai lets your team approve optimizations with one click, which Sedai then safely executes. Once you trust the platform, you can turn on Autopilot mode for resources or environments that you select, which enables Sedai to make optimizations autonomously. Across the board, Sedai enables your team to instantly roll back a change, schedule maintenance windows, & pause optimizations for any reason. ### How does Sedai integrate with our CI/CD pipeline? Does it lead to IaC drift? Sedai is built to work alongside your existing CI/CD & GitOps workflows. Whenever Sedai optimizes a resource, it automatically opens a pull request in your Git repository reflecting the changes, keeping your IaC in sync across Terraform, Helm, & Kustomize. If your pipeline would otherwise overwrite Sedai's changes, our built-in overwrite protection (including a Kubernetes Admission Controller for GitOps tools like Argo CD or Flux) ensures optimizations are preserved, without disrupting your workflow. The result: full auditability, no hidden drift, & no rip-and-replace required. ### What is the difference between an “automated” system and a truly “autonomous” system, like Sedai? This is the #1 most important question to ask when you’re evaluating any system that will make changes in production, without a human in the loop. Automated systems follow rigid instructions & scripts: “if X, then do Y.” Meanwhile, autonomous systems make decisions in real time to achieve big-picture goals, like upholding your SLO for latency when your website traffic spikes. Sedai is able to make these intelligent, contextual decisions, because our platform uses ML to understand the behavior of your apps. For a deeper answer, check out this [in-depth blog](https://sedai.io/blog/automated-vs-autonomous-why-the-difference-matters-for-modern-cloud-operations). ### How do you assess Sedai’s impact on application performance? What metrics does the platform look at? **** Sedai uses a combination of metrics to paint a full picture of your application behavior. This includes the “golden signals” of latency, traffic, errors, & saturation — as well as seasonal trends. Our platform integrates with your existing APM tools, and then it applies patented machine learning to develop a thorough understanding of each app & microservice. ### Do we have to install agents for Sedai to work? Sedai offers both agent-based and agentless deployment options, dependent on your use case. For most cloud resources, Sedai connects securely via IAM permissions, without requiring an agent. However, for Kubernetes clusters, we recommend using our Smart Agent for deeper integration. You can also choose agentless setups for AKS, EKS, & GKE clusters, if you prefer. Onboarding is quick — typically just 15 minutes. ### I have a technical question that isn’t answered here. Now what? You can try asking our website chatbot. But the best way is to [schedule a meeting](http://sedai.io/demo) with our team. This will be a 30-minute conversation with a technical expert, who can address any question about how Sedai works with your infrastructure and tools.