Soft
Main hero

AI Agent

Knowledge assistant

An AI assistant helps employees find and understand approved information while respecting their access rights.

Knowledge Assistant

Company knowledge is often scattered across the corporate portal, product documentation, CMS, CRM, helpdesk, onboarding materials and operational procedures.

An assistant does not need to move everything into a new system. It can help locate the right section, compare information, prepare a summary or draft response, and direct the user back to the original source.

Sources, access and responses

Responsibility layer

Application

Manages authentication, permissions, approved data sources and access boundaries between users and organisations.

Retrieval layer

Retrieves the relevant information, filters it according to permissions, metadata and freshness, and builds a safe context to be passed to the model.

AI model

Analyses the supplied context and uses it to produce a response, summary or other required output. The model should not independently expand the scope of its access to data.

User

Verifies the accuracy and usefulness of the response before using it in important business, operational or organisational decisions.

The model does not enforce permissions itself. The application selects only documents that are approved and accessible to the current user, and then provides the context required for the question.

Indexing does not guarantee that information is either current or complete.

Practical applications

In a hypothetical technical support scenario, the assistant could locate Symfony Messenger documentation and identify sections relevant to a queue-related issue.

For a sales enquiry about modernising a legacy PHP application, it could compare approved materials without creating an unintended commitment regarding project scope.

It could also help a new employee locate onboarding materials or prepare a list of approved deployment procedures.

These are illustrative examples and do not describe GiSoft implementations.

A response containing a citation is useful only when the cited passage genuinely supports the answer and is accessible to the user.

If no approved source is available, the materials conflict, a procedure has been superseded, or the metadata is incomplete, the assistant should make this clear and refer the matter to the owner of the documentation.

The most recent date does not automatically resolve a conflict.

Data, languages and maintenance

There is no need to send the entire knowledge base to the AI provider. The application prepares the sources and metadata, retrieves permitted passages, selects the context, and then checks the structure of the response and its references.

A user may ask a question in one language and receive an answer in another, while retaining access to the original source.

Automatic translation requires particular care with technical, financial and contractual content.

For supporting information, a link and the ability to verify the source are often sufficient.

Operational, financial, legal or production-related recommendations require rules defined by the organisation and the appropriate level of authorisation.

If the assistant or document system is unavailable, an alternative working process suited to the business process is required. Continuous availability should not be assumed.

Measuring and maintaining the solution

An implementation can be evaluated by measuring the time required to find information, the usefulness of sources, the number of corrections and rejected answers, documentation gaps, employee feedback and the cost per query.

The assistant requires ongoing maintenance of its sources, document ownership, versions, access rules, monitoring and cost controls.