Configuration

Models and agents

Match model capabilities to the intended input and task. Review agent configuration separately from model availability.

Memory and knowledge

Choose the appropriate scope for guidance. Verify source readability and retrieval before relying on document-based responses.

Tools and integrations

Enable only the capabilities intended for the deployment. Review external-service boundaries and keep credentials out of conversations and public documentation.

Change with evidence

Record the setting changed and validate the affected workflow. Preserve a rollback path for consequential configuration changes.

Separate personal workflow from administration

Begin with the Agent and Model controls exposed to you. Use the Dashboard to review prompt layers and memory context where available. Ask for an explanation before changing a setting. Controls for models, providers, account management, and system-wide policy may require an administrator.

Ask Friday to guide you

In an authorized FridayLocalAI conversation, try:

Explain the FridayLocalAI setting I am looking at: [label and non-sensitive description]. Tell me its purpose, whether it affects this conversation or a broader scope, and how I can verify a change. Ask for clarification if the label is unfamiliar.

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.

Begin with the scope of a setting

Before changing a setting, ask what it affects. Some choices concern the current conversation or future reply. Others can influence a project, a broader workspace, or the operation of a service. The label alone may not tell you the complete scope, so use the relevant Help information and operator guidance when the effect is unclear. Write down the intended outcome in plain language. For example, you may want future replies in a conversation to use a particular format, or you may need to understand which model was used for a response. Those goals do not necessarily require the same control. Choosing the appropriate mechanism is more reliable than changing several settings and hoping the desired behavior appears.

Model choices and Agent choices

The Model selector represents a requested choice for future replies, while Agent and capability routing can affect the active model. Inspect attribution on the actual result when checking behavior. Changing a requested choice does not rewrite earlier replies. Model-management controls may expose additional options to authorized users, but their presence and permitted use depend on the environment. An Agent represents a role and associated configuration rather than a guarantee of expertise or unrestricted access. Choose an Agent appropriate to the task and test it with a small example. If the role requires a capability that is not available, ask the operator about the configuration. Do not assume an Agent label enables source access, a provider connection, or a Tool operation.

Prompt layers and reusable guidance

Dashboard can surface project, folder, and conversation prompt information where available. Use a prompt layer when the instruction belongs to that part of the work. Keep wording clear enough that you can identify its purpose later. If several layers contain conflicting guidance, review the actual instructions instead of repeatedly adding new corrections to unrelated conversations. Memory Notes are managed separately and have their own scopes and categories. Use them for appropriate durable guidance, and verify the saved note after adding or editing it. Neither a prompt nor a memory record should be treated as a substitute for permissions. Instructions can shape how a task is approached; they do not establish authority to change a file, contact a service, or access unrelated data.

Connections require an explicit understanding

A local-first deployment can include configured external services. Before using a connected capability with sensitive material, confirm what information is sent, which service receives it, and what operating rules apply. Do not infer those details from an icon or a general description of private AI. The specific workflow determines the relevant boundary. Keep credentials in the approved configuration process rather than a conversation or public support message. Ordinary users should not need to paste secret values into chat to understand a feature. If configuration is incomplete, ask the operator for the appropriate next step. An explanation of how a service works is different from evidence that the service is enabled and suitable for the intended task.

Make a small, reviewable change

Change one relevant setting at a time where possible. Record the previous state, the intended change, and the small task you will use to test it. For an important or broad change, follow the operator’s approval and recovery procedure. The user-facing guide should not be read as permission to change system-wide policy or restart services. After saving, inspect the setting in the interface and repeat the test. Check both whether the change persisted and whether it produced the expected behavior. If the result is unexpected, preserve that observation before trying another change. This makes it easier to distinguish an unsaved control, an unrelated context issue, and a misunderstanding about what the setting actually governs.

Keep configuration understandable over time

As a project develops, review the guidance and choices that still affect it. Remove ambiguity from wording, correct outdated requirements through the permitted workflow, and keep a record of why consequential choices were made. A configuration is easier to operate when the people using it can explain the important decisions rather than relying on accumulated trial and error. Ask Friday to help summarize a proposed change or draft a verification checklist, using the non-sensitive details you provide. It should identify unknowns instead of pretending to inspect hidden settings. If the interface or Help wording differs from this guide, use the deployment-specific instructions and report the discrepancy. The goal is a deliberate relationship between the task, the chosen control, its scope, and the result you can observe.