Translate technical findings
Ask what a firewall, encryption, port, endpoint-protection, backup or vulnerability result means in practical language.
100% LOCAL AI ASSISTANCE
Cybersecurity explanations without the cloud.
CSAEC — An integrated AI model designed to run directly on the workstation. Core AI interaction does not require an Internet connection or a cloud AI subscription.
ON-DEVICE INTELLIGENCE
The model works locally, next to the assessment.
The software architecture includes a local language-model service that can use an embedded Qwen2.5 1.5B Instruct model in GGUF format through llama-cpp-python. Depending on the authorised build and local configuration, an optional locally available Ollama runtime may also be supported. In both cases, the purpose is to provide explanations on the workstation rather than sending questions to an external cloud AI endpoint.
This local design means the AI can support a user even when the computer is not connected to the Internet. It does not require a web account, online chat service or recurring cloud AI subscription for its core explanatory function. Availability and response speed depend on the installed build, model initialisation, local storage, memory and processor resources.
The assistant receives the security context selected by the user and returns general educational guidance. Requests are mediated by the local AI service and serialised so that model access remains controlled. The assistant does not independently browse the Internet, transmit a scan report or alter Windows configuration.
LOCAL AI STATUSON DEVICE
Cloud endpoint: none
Internet required: no
Prompt processing: local
Automatic changes: disabled
Human review: requiredWHAT THE AI CAN DO
A guide for understanding, not an autonomous administrator.
Connect risk and impact
Explore why a finding may matter to confidentiality, integrity, availability, recovery or business continuity.
Suggest questions and next steps
Receive general remediation guidance that can be reviewed against policy, vendor instructions and the actual environment.
Support the Security Wiki
Use a dedicated knowledge window while reviewing the main assessment, keeping explanations close to the relevant control.
Avoid automatic cloud submission
Keep core prompts and responses on the workstation rather than transmitting findings to an online AI service.
Work without Internet access
Use the embedded local model in environments where an Internet connection is unavailable, restricted or deliberately disconnected.
HOW TO ASK THE ASSISTANT
Use AI output with professional discipline.
Open a relevant finding
Select a control that is not marked OK and read the original status, explanation and limitation before asking the assistant.
Review the question
The interface can prepare a question based on the finding. Make sure it represents the issue you want to understand.
Read both sources
Compare the response with the structured finding rather than treating the generated explanation as a replacement.
Verify before changing
AI responses can contain errors, omit local constraints or suggest actions that are inappropriate for a production system.
AI CAUTION
Human review is always required.
The local model provides informational assistance only. It does not certify security, diagnose an active incident, guarantee remediation or autonomously modify Windows settings.
Why local AI can create practical value.
Cloud-based assistance may be unavailable because of connectivity, policy, confidentiality or subscription constraints. A local model reduces dependence on an external provider for basic explanations and keeps the workflow available close to the source data. It can shorten the time needed to understand terminology and prepare a focused discussion with qualified personnel.
Local processing is not the same as perfect privacy. Users must still protect the workstation, local files, reports and screen access. Optional email clients, operating-system services or external components can involve third parties when deliberately used.
Technical summary
- Embedded local language-model architecture
- Qwen2.5 1.5B Instruct GGUF model family
- llama-cpp-python local execution
- Optional local Ollama integration
- Controlled, serialised requests
- No cloud AI subscription required
- No Internet required for core local AI
LOCAL AI GOVERNANCE
Keep generated guidance inside a controlled process.
Running the model locally reduces reliance on an online provider, but the organisation should still define who may use the assistant and how its output is handled. Generated text can be copied, saved or shared by a user, so existing confidentiality and acceptable-use rules continue to apply.
The assistant should be used to improve understanding and prepare questions, not to authorise high-impact changes. When guidance concerns firewall rules, encryption, endpoint protection, updates, backup configuration or exposed services, compare it with official vendor documentation and the organisation’s change process.
Local AI performance depends on the workstation. Model initialisation may take time, and resource limitations can affect availability or response speed. The absence of an Internet requirement does not mean that every computer will provide identical performance. Technical requirements should be confirmed for the authorised build.
Organisations can benefit from the predictable data boundary: the core model is on the device, prompts are processed locally and no external cloud AI account is required. This can simplify deployment in disconnected, restricted or confidentiality-sensitive environments while preserving the need for human review.