Gi is the conversational AI assistant employees and managers use on the platform, grounded in your own company knowledge: the documents your admins upload and the content published on your platform.

Gi answers questions about company knowledge, policies, contacts, training and platform use. It answers from the documents, policies and content your organisation makes available to it, and is instructed not to answer beyond them.

When an answer is based on platform content, Gi can include a link to it so users can read the full context. When Gi does not find enough information, it is instructed to say so instead of guessing. Like any AI assistant, it can still make mistakes.

Gi is available on the platform when it is enabled for your organisation.

Gi is the assistant for employees. GFoundry Intelligence also includes agents for the backoffice, each with its own help article:

Employees: find information about the company mission, values, career progression, compliance policies, benefits and other common internal questions.

Managers: locate information for talent management, internal mobility and other people-related questions.

Gi draws on two sources of knowledge inside your GFoundry environment:

Gi's knowledge is split into areas:

An uploaded document goes to the area of the Document Category it is assigned to. Platform content is classified automatically into one area, by its dominant theme. Gi identifies the topic of each question and looks for the answer in the most relevant area.

Because each document lives in a single area, a long master document that mixes several topics will only be found for questions routed to that area. Splitting the content into focused, single-topic documents helps Gi reach the right answer reliably.

Where GFoundry has computed a behavioural profile for the user, Gi can adjust the tone and suggestions of its answer, for example using the user's role, team, whether they manage people and their behavioural archetype. Gi does not add the user's name or email to what it sends to the AI provider, and it is instructed never to mention the archetype to the user.

The way you organise your knowledge has a direct impact on the quality and consistency of Gi's answers.

Prefer short, focused documents over long master files that mix several themes. A document called Vacation and Attendance FAQ will outperform the same content embedded inside a generic Employee FAQ. As a working split for a typical employee FAQ, consider:

A title that names the topic explicitly (for example, Vacation Policy 2026) is easier to find than a generic one (for example, Document 14).

Lead each document with a short summary of its main topic, then go into the detail. This is especially important for PDFs.

Keep one document per language rather than mixing translations in the same file.

Restrict documents to groups only when the content is genuinely confidential or role-specific. Over-restricting reduces the pool of knowledge available to most users.

Outdated documents stay in the knowledge base and can compete with the current source of truth. Archive or update them as part of your normal knowledge-management routine.

Conversations with Gi are stored in GFoundry. Administrators can read individual chat sessions in the backoffice, as a content-quality tool. Let employees know this, for example with a short notice at the chat entry point.

Gi uses external AI providers (OpenAI and Google Gemini, depending on the knowledge area). The provider receives the user's question, the earlier turns of the conversation, the document excerpts relevant to the question and, where available, the profile context described above. Keep in mind that document excerpts are sent as written: if your documents contain names, emails or phone extensions, for example an internal contacts directory, those excerpts reach the provider when they are used to answer.

Group-based document visibility is applied when Gi searches for an answer. A document restricted to specific groups is not used to answer a question from a user outside those groups.

The Data Processing Agreement, including the list of sub-processors, is available on request. For requests about stored conversations, contact GFoundry support.

Admins have a Gi Bot area in the backoffice, with Document Management and Chat Sessions.

Select New Document, give it a Document Name, use Upload File, choose the Visibility and the Document Category, and optionally set Groups Segmentation. The visibility options are:

After you save, the document shows Waiting for indexing... and then Indexing... until it is Published. If indexing fails, the status is Error.

Chat Sessions lists the conversations with their first message, the date and the number of interactions. Open a session to read its Chat History.

Each document has a History section that records changes such as file updates, name and category changes, status updates and group segmentation changes.

Rephrasing changes how the question ranks against the available documents. If you see this on a recurring topic, the source document is usually mixed with other themes and would benefit from being split into a focused single-topic document.

No. Each document belongs to a single area. If a document covers several topics, split it into separate single-topic documents.

Uploaded documents are added by admins in the backoffice and typically contain reference material such as manuals, policies and FAQs. Platform content is the content already published on your GFoundry platform, which Gi learns from only where the content sync is set up for your organisation.

An uploaded document is available once its status is Published. Where the platform content sync is set up, it runs once a day.

Once the new version is indexed, Gi uses it for the next questions. There is no need to clear conversations or restart Gi.

Within a chat session, Gi uses earlier turns as context. Earlier sessions are stored so admins can review them.

Treat it as a signal that the knowledge base needs work. The most common causes are an outdated document still present in the corpus, two documents disagreeing on the same topic, or a master document where the relevant section is buried under unrelated content.

Open a support ticket with the date and time of the conversation, the question, the answer received and the answer expected. The support team can then trace which documents were used and recommend the right correction.