Akamas
Akamas is an autonomous optimization platform that tunes Kubernetes, cloud infrastructure, application runtimes and GPU inference stacks together. Its patented reinforcement-learning engine turns telemetry you already collect into reviewable configuration changes that cut cost without trading away reliability.
What is Akamas?
Akamas is an autonomous optimization platform for Kubernetes and cloud-native applications, built by Akamas S.p.A. in Milan as a spin-off of the performance-engineering firm Moviri. Its premise is that observability, CI/CD and infrastructure automation have all matured while configuration tuning stayed manual, reactive and split across teams — a gap the vendor describes as a missing platform capability rather than a tooling problem.
What distinguishes it is the breadth of what it tunes at once. Most tools work on a single layer: pod right-sizing, instance selection, or engine parameters. Akamas analyses infrastructure, Kubernetes objects and application runtimes together, because the interactions between them are where efficiency and reliability actually come from. In practice that means pod requests and limits alongside JVM heap and garbage collection, Node.js V8 settings alongside container limits, HPA and KEDA baselines, node-group instance families, Spark job and cluster settings, and — since June 2026 — GPU resources, serving engines and inference runtime parameters for AI workloads. The engine behind it is proprietary reinforcement learning, which the company presents as patented and peer-reviewed rather than heuristic.
The platform ships in two shapes sold under one licence. Akamas Insights works on live production data: it connects to an existing observability stack — Prometheus, Amazon Managed Prometheus, Datadog, Dynatrace or Grafana Cloud — with no agent and no code change, then surfaces ranked recommendations with an estimated impact. Akamas Offline works in a test environment, designing, running and scoring performance experiments until it converges on an optimal configuration; it is an on-premise product installed on the customer's own machines via Docker Compose, Kubernetes or OpenShift, and driven by a CLI.
Autonomy is deliberately bounded. Changes can be applied automatically or, more commonly, delivered as Git pull requests carrying projected savings and latency effects, compatible with Argo CD and Flux, so every change stays versioned, auditable and reviewable. Tuning profiles let a team push aggressive cost targets on development namespaces while keeping conservative safety margins on critical services. Akamas names Sisal and Sabre among its customers and appears in the GigaOm Radar for Cloud Resource Optimization 2026.
What it does
- Analyse Kubernetes clusters, workloads and nodes to locate wasted resources and reliability risks
- Recommend right-sized CPU and memory settings for pods, node groups and application runtimes
- Tune JVM heap and garbage collection, and Node.js V8 settings, jointly with pod limits
- Optimise the resource baseline that HPA and KEDA autoscalers scale from, without touching scaling thresholds
- Deliver each change as a reviewable Git pull request carrying projected savings
- Design, run and score automated performance test campaigns until an optimal configuration is reached
- Tune GPU, serving engine and inference runtime parameters against a single latency objective
When to use Akamas / When not to
A quick filter to help you decide if Akamas is the right fit.
When to use Akamas
- Platform engineering teams running large Kubernetes estates who need cluster efficiency and autoscaling behaviour under control
- SRE and DevOps teams that want misconfigurations caught before they turn into incidents
- Organisations whose cloud bill is dominated by Java workloads, where JVM heap and garbage collection defaults were never meant for containers
- Performance engineers who must find an optimal configuration in a test environment before a release ships
- Teams running GPU inference in production and trying to lower cost per token while holding a latency target
When not to use Akamas
- Buyers who need a published price: Akamas quotes on request and shows no figures anywhere on the site
- Teams without an existing observability stack, since Akamas reads Prometheus, Datadog, Dynatrace or Grafana Cloud rather than collecting metrics itself
- Anyone expecting a mobile app: there is no iOS and no Android application
- Organisations that require a certified supplier, as the vendor states it holds no SOC 2, PCI-DSS or HIPAA certification
- Teams unwilling to run software in-house, because Akamas Offline is an on-premise product to install and operate
How to use Akamas
A typical end-to-end flow, from setup to results.
- Create an Akamas Insights account at the vendor's sign-up page and wait for the personal environment to be provisioned, which the documentation times at around three minutes
- Connect an observability source: Prometheus, Amazon Managed Prometheus, Datadog, Dynatrace, Grafana Cloud, or an on-premise extraction
- Let the platform read cluster, workload and runtime metrics — no agent is deployed and no application code changes
- Work through the cluster overview and the ranked recommendations, each carrying an estimated impact
- Set tuning profiles so development namespaces and tier-one services receive different optimisation targets
- Apply a recommendation directly, or push it to your Git repository as a pull request for developer review
- For pre-production work, request a sandbox of Akamas Offline or install the product yourself
- Install Akamas Offline via Docker Compose or on a Kubernetes or OpenShift cluster, online or from a private registry, then install the CLI and the licence file
- Model the system, its components, a workflow, and the goals and constraints of the optimisation
- Launch an optimisation study and let Akamas iterate through performance experiments until it converges
Pros & Cons
Pros
- Tunes infrastructure, Kubernetes and application runtimes together, which few competing tools attempt
- Public, versioned technical documentation covering both products, including installation and reference material
- Unusually transparent security page listing what the platform does not collect, and stating plainly that it holds no security certification
- Agent-free setup for Insights: it plugs into observability you already run, with no code change
- Every change stays reviewable and reversible through Git pull requests before it reaches production
- Free trial with automatic environment provisioning, plus a free cluster assessment and an Offline sandbox
- Independent analyst recognition in the GigaOm Radar for Cloud Resource Optimization 2026
Cons
- No public pricing whatsoever: every figure has to come from a sales conversation
- Unusual billing unit — Optimization Credits — whose consumption depends on managed components and packs in use
- No security certification at all, as the vendor itself states: no SOC 2, no PCI-DSS, no HIPAA
- No hosting country or region disclosed for the platform
- The only published privacy policy covers the marketing website, not the product
- Akamas Offline has to be installed and operated on-premise by the customer
- English only, with no published stance on whether customer telemetry feeds model training
Pricing & Plans
Akamas publishes no prices. The pricing page states that the model depends on what is being optimised and on the scale of the environment, and invites prospects to describe their setup so that figures can be worked out together. Contractually, the metering and billing unit is the Optimization Credit, consumed while an optimisation study is running and priced according to the contract term set out in the order document; the number of credits in use depends on the optimisation packs, integration packs and managed components involved. There is no permanent free plan, but a free trial of Akamas Insights, a free cluster assessment and an Akamas Offline sandbox are all offered at no cost.
- Akamas Insights free trial — self-service account with an automatically provisioned environment
- Free Assessment — a personalised review of a Kubernetes cluster's savings potential
- with no credit card and no commitment
- Akamas Offline sandbox — a trial environment with sample data
- available on request
- Commercial licence — governed by a Licensing Framework Agreement plus an order document setting the products
- services
- Optimization Credits
- term
- site
- price and territory
Data, GDPR & hosting
A consolidated view of how Akamas handles your data.
GDPR overview
GDPR is handled contractually rather than through a compliance badge. Article 16 of the Licensing Framework Agreement binds both parties to Regulation (EU) 2016/679, limits Akamas and its subcontractors to the processing strictly required by the agreement, imposes technical and organisational measures, and gives the customer a written audit right at reasonable intervals. Transfers outside the EEA are allowed without prior written consent only where the destination country offers an adequate level of protection and the obligations of the DPA are satisfied. Italian law governs, with exclusive jurisdiction in Milan. The website privacy policy, last updated on 6 July 2026, carries a dedicated section for European Union users with access, rectification, erasure and portability rights exercisable by e-mail. Mind the scope: that policy describes the website, not the platform.
Who owns the data?
The published privacy policy names Akamas S.p.A., Via Schiaffino 11, Milan, as owner and data controller, reachable at support@akamas.io — but it covers the akamas.io website only, not the platform. For the product itself, the Licensing Framework Agreement leaves the customer in control: Akamas processes personal data supplied to it strictly as required to perform the agreement, and undertakes not to disclose it to third parties unless the customer requires otherwise. What the platform ingests from a customer environment is resource and configuration telemetry — pod metrics, node capacity, runtime settings — which the vendor states contains no application content and no business logic.
Reuse rights
The Licensing Framework Agreement is a software licence, not a data licence: it grants the customer no new rights over data because the customer already holds them, and Akamas acts only on what it is given. The licence running the other way is narrow — limited, non-exclusive, non-transferable and non-sublicensable, bounded by the contract term, the territory and the number of Optimization Credits purchased. Optimization findings are delivered inside the customer's own workflow, including as Git pull requests that stay versioned in the customer's repository, so recommendations can be reused, audited or reverted freely. No clause was found allowing Akamas to reuse customer telemetry for its own purposes — and none was found forbidding it either, which is worth noting rather than assuming.
Data retention & training
Hosting summary
No hosting country or region is published for the platform, and that gap is worth stating plainly rather than filling in. Akamas Insights is a hosted service: the vendor provisions a personal environment, authentication runs through Auth0 on OAuth 2.0 and OpenID Connect, and access is through the browser. The security documentation commits to HTTPS with TLS 1.2 or higher in transit, encryption at rest for metrics data and backups, and encrypted storage of integration credentials — but it names no data centre, no cloud provider and no jurisdiction. Akamas Offline sits at the other extreme: it is an on-premise product running on a machine inside the customer environment, whether a data-centre server, a cloud virtual machine or a laptop, so the hosting question is answered by the customer's own choices. The Licensing Framework Agreement permits transfers of personal data outside the European Economic Area where the destination country offers an adequate level of protection. The processing locations that do appear publicly — the United States, Germany, Italy and Israel — belong to the marketing website's third-party services and should not be read as the platform's hosting footprint.
Things to keep in mind
Risks and trade-offs to weigh before adopting Akamas.
- The headline gains — up to 60% lower cloud cost, 30% better performance, 58% on JVM workloads — are vendor claims measured in vendor conditions, not audited results; treat them as an upper bound
- The published privacy policy covers the akamas.io website, not the platform, so most GDPR facts a reader can verify do not describe the service they would be buying
- No hosting country or region is disclosed for the platform, and no subprocessor list is published for the product itself
- Handing configuration decisions to an autonomous system can erode a team's own understanding of why a setting is what it is; the pull-request workflow only helps if the reviews are real
- Annex A of the licence agreement is dated boilerplate: it cites partners under former names and points to an archived February 2025 snapshot of the site for the current description
- The vendor states it holds no security certification, which is honest but consequential for regulated buyers
- The last two published financial years at the Italian business register show operating losses, which is normal after a first institutional round but worth factoring into a long-term commitment
Setup & Integrations
Technical difficulty
Two very different answers. Akamas Insights is light: create an account, connect an observability source, and the environment is provisioned automatically in about three minutes, with no agent and no code change — though it assumes you already run Prometheus, Datadog, Dynatrace or Grafana Cloud and can grant read access to your clusters. Akamas Offline is heavy: hardware, software and network prerequisites, installation through Docker Compose or on a Kubernetes or OpenShift cluster, a CLI to install and initialise, a licence file to apply, and a system, workflow, goals and constraints to model before the first study runs.
Deployment
Integrations
Supported languages
Behind Akamas
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What exactly does Akamas optimise?
Do I need to install an agent or change my code?
Which data sources does it read?
Does Akamas change my configuration on its own?
Is there a free trial?
How much does it cost?
Is there an API?
Is Akamas SOC 2 or ISO 27001 certified?
Where is the data hosted?
Is there a mobile app?
Should you pick Akamas?
Akamas occupies a narrow but genuinely differentiated position. Where most cloud-efficiency products stop at infrastructure right-sizing or cost reporting, Akamas tunes infrastructure, Kubernetes and application runtimes as one system — and that is the part competitors rarely attempt. The engineering lineage is real: two decades of performance work at Moviri, proprietary reinforcement learning presented as patented, a $10 million round led by United Ventures in February 2025, and an appearance in the GigaOm Radar for Cloud Resource Optimization 2026.
The product is also unusually candid in places where vendors normally are not. Its security documentation lists what the platform does not collect — application logs, database content, Kubernetes secrets, network traffic — and states plainly that Akamas Insights holds no SOC 2, PCI-DSS or HIPAA certification. That is a commercially costly sentence to publish, and it is worth more than a badge without a certificate number.
The reservations are equally concrete. There is no published price at all: the model runs on Optimization Credits whose unit cost depends on contract term and on how much of your estate you optimise, so budgeting means talking to sales first. No hosting country or region is disclosed for the platform, and the only privacy policy published covers the marketing website rather than the product — so most GDPR facts a reader can check do not actually describe the service. Akamas Offline must be installed and operated in-house. The site is English only, and there is no mobile application.
For a platform, SRE or performance-engineering team already running observability at scale, Akamas answers a question those tools do not: not what is happening, but what the configuration should be. For anyone wanting self-service pricing or a certified vendor, it is the wrong door.
- Choosing a selection results in a full page refresh.
- Opens in a new window.