A single investigation can fit comfortably inside a standard service envelope. Parallel agents, repeated tool calls, long contexts and sustained workflows can create a very different throughput profile.
CSI PRO Enterprise is designed for that transition. It extends the managed operating model around Cybersecurity Superintelligence (CSI) with substantially more capacity while keeping the infrastructure operated by Alias Robotics.
When standard PRO no longer matches the operating envelope you need
Standard CSI PRO has a published maximum of 500,000 tokens per minute (TPM) and 60 requests per minute (RPM). Alias Robotics’ current PRO terms describe PRO access within the Alias-hosted operating model.
For many teams, that is the right operating model. The workload fits the standard limits, Alias-hosted inference is acceptable and there is no architectural requirement to take control of the underlying inference infrastructure.
At larger scale, the constraint can change. A security organisation may want to run more agentic investigations in parallel, support higher peaks, maintain sustained automated workflows or expand a deployment across more security processes. In those cases, the question is not whether standard PRO is “enough” in the abstract. It is whether its published throughput matches the organisation’s actual operating envelope.
Pricing, licensing, included-token entitlement and procurement questions remain the job of the separate CSI PRO Pricing FAQ. CSI PRO Enterprise addresses a narrower question: what if the organisation needs materially more managed capacity?
Enterprise capacity from ×5 to ×25
All CSI PRO Enterprise tiers are annual plans. Capacity scales from five to twenty-five times the standard PRO TPM and RPM envelope.
| Tier | Capacity | TPM | RPM | Data treatment |
|---|---|---|---|---|
| Standard CSI PRO | — | 500,000 | 60 | — |
| Enterprise Platinum | ×5 | 2,500,000 | 300 | Opt-out |
| Enterprise Ultimate | ×10 | 5,000,000 | 600 | Opt-out / ZDR |
| Enterprise Titanium | ×15 | 7,500,000 | 900 | Opt-out / ZDR |
| Enterprise Diamond | ×20 | 10,000,000 | 1,200 | Opt-out / ZDR |
| Enterprise Elite | ×25 | 12,500,000 | 1,500 | Opt-out / ZDR |
The multiplier refers only to the TPM and RPM capacity shown. It does not represent users, seats, simultaneous agents or any other commercial entitlement.
TPM and RPM measure different kinds of load
Token-heavy workflows
Agents carry large contexts, iterate through reasoning and tool-use loops, or exchange substantial intermediate state.
Request-heavy workflows
Many shorter model calls generated across parallel tasks.
Increasing both limits proportionally expands the managed operating envelope without turning either metric into an unlimited resource.
Agentic cybersecurity makes this distinction practical. Alias Robotics’ CSI harness research evaluated a blackboard architecture in which scaffold-specialised agents ran in parallel and exchanged intermediate findings. In the underlying paper, the 33-challenge Cybench evaluation reports 1.08 billion input tokens and 18.9 million output tokens for that architecture.
That benchmark does not define a production sizing rule. It illustrates the scaling dynamic: multi-agent execution can create model traffic far beyond a single prompt-response interaction.
Enterprise scale without operating the infrastructure yourself
Higher throughput and infrastructure control are separate decisions.
CSI PRO Enterprise is for organisations that need the former without necessarily wanting the latter. Alias Robotics operates the infrastructure, while the customer selects an Enterprise capacity envelope aligned with its workload. That preserves the managed-service model while creating substantially more room for sustained or bursty Cybersecurity AI operations.
This distinction matters because operating a customer-controlled AI stack is not merely a licensing choice. It introduces infrastructure, deployment, lifecycle and operational responsibilities that some organisations explicitly need, but others would rather not own.
A large security team can require higher throughput without having a sovereignty or air-gap requirement. Conversely, a smaller deployment can require On-Premise operation because infrastructure and data control are architectural constraints.
CSI PRO Enterprise exists between those requirements: more managed capacity without making customer-operated infrastructure a prerequisite.
Data treatment for enterprise requirements
Capacity is only part of the Enterprise decision. Data treatment can be a separate requirement.
Opt-out
Customer data are excluded from training or model-improvement uses. Alias Robotics’ current PRO terms distinguish exclusion from training from the processing necessary to provide and maintain the service.
Zero Data Retention (ZDR)
A stronger retention control: customer data are not retained, subject to applicable legal and regulatory requirements. Available on Ultimate, Titanium, Diamond and Elite.
Zero Data Retention is available on eligible Enterprise tiers where permitted by applicable legal and regulatory requirements.
Opt-out and ZDR should therefore not be treated as synonyms. One governs whether customer data can be used for training or model-improvement uses; the other addresses retention. Neither should be inferred from the capacity multiplier alone.
CSI PRO Enterprise or CSI On-Premise?
The key decision is not “which tier is higher?” It is which operating requirement is driving the architecture.
| Requirement | Standard CSI PRO | CSI PRO Enterprise | CSI On-Premise |
|---|---|---|---|
| Inference / infrastructure operating model | Alias-hosted model access | Alias-operated infrastructure | Customer-controlled, on customer hardware |
| Defined throughput envelope | 500,000 TPM / 60 RPM | 2.5M–12.5M TPM / 300–1,500 RPM | No standard public TPM/RPM figure |
| Higher managed throughput | Standard envelope | ×5 to ×25 | No standard managed-throughput tier published |
| Customer-controlled inference infrastructure | No, for Alias-hosted model access | No | Yes |
| Air-gapped / sovereign deployment | Not the standard hosted path | Not the managed-service objective | Publicly supported for On-Premise |
| Custom datasets, fine-tuning, bespoke agents, steering, private benchmarking | Not part of the standard published PRO entitlement | Not specified by the Enterprise capacity offer | Publicly described for On-Premise |
Alias Robotics’ current On-Premise offer is explicit about the architectural difference: it describes CSI running on customer hardware, with air-gapped Alias models, custom fine-tuning, bespoke agents, steering and private benchmark suites.
The Sovereign AI for National Cyber Defense case study documents one European national defence deployment running CSI entirely on-premise while retaining control over infrastructure, models, operational data, deployment lifecycle and AI governance.
That is a different requirement from managed scale. CSI PRO Enterprise is not a halfway version of sovereignty, and CSI On-Premise is not simply the next pricing tier after Elite. See CSI PRO vs On-Premise for a deeper look at trust boundaries.
Choosing the right capacity tier
The right Enterprise tier should be sized from workload behaviour, not company size.
- Sustained TPM: start with observed or expected sustained tokens per minute.
- Peak TPM: model peaks rather than relying only on averages.
- RPM separately: workloads composed of many short calls can behave differently from workflows dominated by fewer, context-heavy requests.
- Agentic workflows: map those rates to the number and behaviour of agentic workflows you expect to run, and leave room for realistic growth.
- Non-throughput constraints: does the organisation need opt-out only, or is ZDR required? Is Alias-managed infrastructure still acceptable?
If the answer to the last question is no because the requirement is sovereignty, air-gap, customer-controlled inference, bespoke models or agents, private datasets, fine-tuning, steering or private benchmarking, the sizing discussion has become an On-Premise architecture discussion instead.
The ×5-to-×25 scale is therefore a capacity framework, not a proxy for organisation size, user count or agent count.
Match capacity to the operating model
Moving Cybersecurity AI into sustained enterprise operations forces two questions that are easy to conflate: how much throughput does the workload need, and who must control the infrastructure?
- CSI PROFits workloads that operate within the published managed envelope of 500,000 TPM / 60 RPM.
- CSI PRO EnterpriseExpands that envelope substantially, from ×5 to ×25, while keeping infrastructure operated by Alias Robotics.
- CSI On-PremiseFor cases where control, sovereignty or bespoke deployment requirements shape the architecture itself.
Frequently asked questions
What is CSI PRO Enterprise?
CSI PRO Enterprise is a set of annual plans from Alias Robotics that scale the standard CSI PRO throughput envelope from ×5 to ×25 while Alias Robotics continues to operate the infrastructure. It is designed for organisations that need substantially more managed Cybersecurity AI capacity without running a customer-controlled On-Premise inference environment.
What are the CSI PRO Enterprise tiers?
Enterprise Platinum (×5, 2,500,000 TPM, 300 RPM), Ultimate (×10, 5,000,000 TPM, 600 RPM), Titanium (×15, 7,500,000 TPM, 900 RPM), Diamond (×20, 10,000,000 TPM, 1,200 RPM) and Elite (×25, 12,500,000 TPM, 1,500 RPM). Standard CSI PRO is 500,000 TPM and 60 RPM.
Does the multiplier include more users or agents?
No. The multiplier refers only to tokens-per-minute and requests-per-minute capacity. It does not represent users, seats, simultaneous agents or any other commercial entitlement.
Which tiers include Zero Data Retention?
Zero Data Retention (ZDR) is available from Enterprise Ultimate upwards (Ultimate, Titanium, Diamond and Elite), where permitted by applicable legal and regulatory requirements. Enterprise Platinum includes opt-out, which excludes customer data from training or model-improvement uses.
Is CSI On-Premise the tier above Enterprise Elite?
No. CSI On-Premise is a different operating model, not a capacity tier. It is for requirements such as sovereignty, air-gapped deployment, customer-controlled inference, bespoke models or agents, private datasets, fine-tuning, steering or private benchmarking.
If standard CSI PRO no longer matches your operating envelope, talk to Alias Robotics about the Enterprise capacity tier that fits your workload. If your requirement is driven by sovereignty, air-gap or customer-controlled infrastructure, explore CSI On-Premise instead.