How FridayLocalAI Works

1. Choose the workspace

Open the intended project, folder, and conversation. State the goal, constraints, and desired output.

2. Select suitable capabilities

Choose an available agent and model appropriate to the task. Check permissions and whether the required tool or integration is configured.

3. Verify the context

Attach supported material and confirm that its contents can be used. Keep durable guidance in the appropriate scope.

4. Review the outcome

Read the response, check relevant source material, and inspect any output artifact. Treat claims of completed actions as claims until their results are verified.

A first ten-minute task

Start with a harmless example: three meeting notes that you type yourself. Ask for a summary and next actions. Check that the answer adds no invented dates or owners. Give one correction, review the revision, and return to the same conversation later to check that the exchange is saved.

Ask Friday to guide you

In an authorized FridayLocalAI conversation, try:

Guide me through summarizing three practice meeting notes in FridayLocalAI. Help me choose the conversation, enter the notes, review the summary, and verify saved history. Pause after each step.

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.

Define a result before starting

Choose an outcome you can recognize: a short summary, an outline, a comparison, or an answer grounded in a source. Explain the audience and any constraints that would change the result. If the request is still vague, ask Friday to help clarify it before producing a long response. A clear goal gives both the assistant and the user a basis for deciding whether the work is useful. For a first example, type three fictional meeting notes and ask for next actions without invented owners or dates. You can compare every output item with the input. This is a simple way to learn the workflow because it does not depend on outside information or a complicated capability. Once the pattern is familiar, apply the same discipline to more substantial work.

Select the context that belongs to the task

Open the intended project and conversation in Organizer. Check the active title and context before sending material. If you are continuing earlier work, review the recent exchange and state what changed. If the task is new or substantially different, a separate conversation may make it easier to understand and find later. Provide only the context needed for the current request. A larger amount of background is not automatically better if it obscures the important question. Use an attachment when the source should be reviewed as a document and the workflow supports it. Use a permitted Memory Note for durable guidance that belongs in an appropriate scope. Keep temporary task details in the conversation when that is sufficient.

Understand the response components

An Agent provides a configured role, while a model produces the response. The requested Model selector may be affected by Agent and capability routing, so inspect the active attribution when evaluating the result. Skills can guide recurring methods, and Tools expose operations governed separately by policy and permissions. That means the assistant can explain an action without being able to execute it. If you need a draft, request a draft. If you need an actual operation, identify the target and the result to verify, and work within the permitted capability. Do not treat a natural-language request as proof that access exists or that the runtime can perform every step implied by the wording.

Check source availability before analysis

For document work, confirm that the contents relevant to the question are available. An upload or preview alone is not proof of model-readable context. Ask a narrow question with an answer you can check in the original, such as a short passage reference. If the content is unavailable, resolve that limitation before requesting a broad interpretation. Once the source check succeeds, ask for the actual analysis and the supporting references. Review not only the quoted material but also the conclusions drawn from it. A correct quotation can still be paired with an unsupported inference. Tell Friday to distinguish source statements, interpretation, and unanswered questions so the output remains useful even when the evidence is incomplete.

Review, correct, and continue

Read the response against the original goal. Check whether it followed the constraints, preserved important facts, and acknowledged missing information. Give a specific correction when needed. A focused instruction such as “keep the original order and remove the unsupported deadline” is easier to act on than a general request for a better answer. If the task involves an artifact or saved change, inspect that result directly. Do not stop at a message saying the work is finished. A generated plan, an attempted action, and a verified completion are separate outcomes. When the result is acceptable, note the decision or next step in the conversation so that future work begins from an understandable state rather than an assumption about what happened.

Return to the work with continuity

Use meaningful names and the Organizer search to find the conversation later. Confirm saved history and review the most recent decision before continuing. If the exchange is long, ask for a summary of facts, decisions, open questions, and next steps, then check the summary against the actual conversation. For recurring preferences, use Memory Notes intentionally and verify the saved scope. For a new question, provide the context that still applies rather than assuming the assistant receives everything from every previous task. FridayLocalAI works best as a deliberate cycle: define the goal, establish context, use available capabilities, inspect the result, and carry forward the parts that have been checked. The application supports that cycle, while human review determines whether the output is ready for its intended use.