AI Governance

Roles and permissions

Controls and server-side gates separate ordinary use from administration and capability changes. An agent cannot grant itself authority.

Scoped context

Global, organization, project, folder, and conversation memory address different levels of context. The scope of a note matters as much as its content.

Governed actions

Skills and tools are subject to configured policy. Changes to capabilities, integrations, and data access should be explicit and reviewable.

Evidence over assertion

Deterministic governance does not mean deterministic or infallible model answers. Verify significant claims and completed operations against the actual result.

Use help without granting new authority

An explanation, a Skill, and an enabled Tool serve different purposes. Ask what an action will affect before approving it. When Friday reports a change, inspect the resulting file, saved record, or visible status. If evidence is missing, treat the result as unverified.

Ask Friday to guide you

In an authorized FridayLocalAI conversation, try:

Before helping with this task, explain what you can guide me through and what would require a permitted tool or operator action. Identify any data changes or external destinations, and ask before consequential changes.

Tell Friday what you can see and which step you have reached. If the instructions do not match your screen, say so; ask for the relevant Help topic or contact your operator. Friday cannot be assumed to see your screen or know your current settings.

Governance starts with ordinary decisions

AI governance is often discussed in broad terms, but users encounter it through everyday questions: which workspace is active, which source is being used, who may change a setting, and what evidence shows an action succeeded. FridayLocalAI’s approach makes those distinctions part of the application rather than leaving them entirely to the wording of a prompt. Begin a task by describing the desired result and its limits. If the request involves changing data or contacting an external service, identify that consequence before proceeding. If it only requires an explanation or a draft, say so. A clear task boundary makes it easier for the user, the application, and the operator to understand what is intended and what would require a separate decision.

Role, scope, and capability have different jobs

A role affects which actions and controls a user can access. Scope identifies the work or context to which information applies. A capability concerns a permitted operation. These concepts interact, but none should be reduced to a friendly Agent name or a statement that the assistant is “allowed to do everything.” The runtime must enforce the actual boundary. For example, a conversation-specific preference should not automatically become organization-wide guidance. A document visible in a permitted workspace should not imply access to unrelated storage. A Skill describing a workflow should not create the Tool permission needed to execute it. Ask Friday to explain the distinction, then inspect the relevant controls or consult the operator when you need a change.

Treat source material according to its purpose

A document can contain facts, examples, quotations, or instructions written by someone else. Its content is not automatically an instruction from the user. When asking Friday to analyze a source, state the task clearly: summarize, compare, extract, or explain. Keep the requested operation separate from any directive that happens to appear inside the material. Likewise, Help information explains application behavior but does not establish that the current service is available. A Memory Note can preserve guidance but does not prove a file was read. A generated summary can aid review but should not replace the original evidence for an important claim. Identifying the purpose of each source helps prevent context from being mistaken for authority.

Make review proportional to the result

A rough outline may need a quick editorial check. A factual document summary needs comparison with the source. A claim that a setting changed needs inspection of the saved setting. A generated artifact needs to be opened and assessed for the intended use. Choose the review method based on what is being claimed rather than applying the same checklist to every response. Ask for assumptions and unresolved questions when they matter to the decision. If Friday cannot establish a fact, a useful answer can explain what evidence would resolve it. That is preferable to a confident answer with no support. Do not confuse a smoothly written response with a verified operation, and do not treat the absence of an error message as complete proof of success.

Data security and sovereignty questions

Searches for AI data security or sovereign AI often reflect a desire for more control over information and infrastructure. Translate that desire into specific questions: where does processing occur, who operates the environment, what connections are enabled, and who can access saved material? Those questions are more actionable than assuming a local model answers every concern. This website does not certify compliance with a particular legal regime or promise that a deployment satisfies every data-sovereignty requirement. Such conclusions depend on the actual configuration, contracts, jurisdiction, and organizational needs. FridayLocalAI can help structure the discussion and keep the technical boundaries visible. The responsible people must evaluate the arrangement and obtain the appropriate advice or evidence for their circumstances.

Respond to a concern without making it worse

If a result appears to expose unrelated data, claims an action that cannot be verified, or behaves outside the intended scope, stop that workflow and preserve a concise account of what happened. Record the time, feature, expected behavior, and sanitized symptom. Avoid spreading private contents or trying multiple disruptive changes before the cause is understood. Use the project’s support route or your operator’s process. Friday can help organize a report if conversation remains available, but the report should distinguish observations from guesses. After a repair, repeat a bounded version of the original task and check the result. Governance is useful when it supports clear decisions, understandable evidence, and an accountable path for correcting problems.

Turn a principle into a review step

Use a small task to make governance concrete. Name the owner, intended outcome, allowed sources, and the action that needs approval. Ask FridayLocalAI to draft a review checklist and identify assumptions, then compare its checklist with the actual project requirements. Record what was accepted and who made the decision. A book or generated answer can help explain a principle; evidence from the installation is still needed to establish that a control operates as intended. The Phronesis Intelligence reading guide explains how these themes connect to practical wisdom, measurement, and source discipline.

Explore the publications and annotated reading guide.