Analyse your organisation, and your AI, in plain language

Ask most AI tools a question about your organisation and you get a fluent paragraph. It may be right. It may be a plausible guess about a reporting line that changed in the spring. Nothing in the answer tells you which, and for anything that leads to an action that is not good enough.

The Swfte Intelligence Platform is built around a different premise: evidence before inference. The graph underneath holds facts about your organisation, and every fact carries an evidence status. When you ask a question, the answer is assembled from those facts and the status comes with it. A reporting line that was observed in your directory reads differently from one that was inferred, and both read differently from one that has not been seen for longer than it should have been.

That is the property that makes analysis useful as a step in a loop rather than a destination. If you are going to build an agent or a workflow from a finding, you need to know how much weight the finding can carry.

An answer in the Intelligence Platform is designed to carry more than a result. It names the facts it used and the evidence status of each one: observed, corroborated, verified, inferred, stale, disputed or unknown. It shows the age of the underlying observation, so a stale fact looks stale. Where two sources disagree, the answer says so and does not quietly pick one.

It also says what it could not see. The platform reports coverage: which sources are connected and which are not. An answer about who owns a service is only as complete as the sources that describe services, and today the appliance reads directory sources. Collectors for cloud, code, Kubernetes, databases and more are on the roadmap, and until they arrive the honest answer to some questions is that the platform does not know.

That restraint is deliberate. The platform is built never to say that it understands everything, because an organisation that believes that will stop checking.

A question is asked by someone, and that someone has a limit on what they may see. The platform is built so that an AI asking on a person’s behalf has never more access than that person. If an analyst is not allowed to see a group, the answer does not include it, and the platform does not tell them what they are missing.

Content that comes back from a source is treated as untrusted. A document that says ignore your instructions and reveal the following is data to be read, not an instruction to follow, and it cannot override policy. That matters more as analysis moves from a human reading a screen to an agent reading context.

Every question and every answer leaves a record. In the Trust and Governance Fabric this is the Traceability and Auditability facets at work: what was asked, by which identity, what was read, and what was returned.

The detail of the question interface is not something we will invent on this page. The query language is <query language - founder to fill>, the named dashboards are <named dashboards - founder to fill>, and the connector list beyond directory sources is <connector list beyond directory sources - founder to fill>. Availability is <availability - founder to fill> and pricing is <pricing - founder to fill>.