Memory & Context

Choose the appropriate scope

Use global and organization guidance only when appropriate and permitted, project guidance for an ongoing area of work, folder context for related material, and conversation context for the specific discussion.

Write intentionally

A conversation mentioning a fact is not, by itself, proof that a durable memory was written. Confirm saved notes through the management interface and the relevant scope.

Keep sources separate

A memory note, a document in storage, and context supplied to a model have different purposes. Review provenance and retrieval when an answer depends on source material.

Review and maintain

Keep guidance current and appropriate to the audience and task. The model’s response does not establish persistence or future access.

Save and maintain useful guidance

Open Dashboard, find Memory, and choose Manage to open Memory Notes. Where permitted, choose Scope and Category, enter a concise note, and use Add. Confirm the saved note in the list. Existing notes can be edited where permitted; changes are versioned. Keep temporary task details in the conversation rather than making every remark durable.

Ask Friday to guide you

In an authorized FridayLocalAI conversation, try:

Help me draft a short Memory Note for this preference: [preference]. Recommend the narrowest suitable scope, explain who or what work it should apply to, and guide me through saving it. Do not say it is saved until I verify the note in Memory Notes.

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.

Steps to follow

1. Choose Conversation for guidance limited to one discussion.
2. Choose Folder or Project when the guidance applies to the related work in that scope.
3. Organization and Global are broader scopes; use them only when your role and the intended audience make that appropriate.
4. Select an appropriate Category such as Preference, Workflow, Fact, or Glossary.
5. After adding or editing, read the saved note and its scope back in the interface.

Decide what should last

Memory Notes are useful for guidance that should remain available beyond one exchange. Examples include a writing preference, a project convention, or a definition used repeatedly in related work. Temporary details, tentative guesses, and one-off instructions may be better left in the conversation. Saving everything creates more material to maintain and can make it harder to identify what still applies. Before adding a note, ask whether it is accurate, stable enough to reuse, and relevant to the intended scope. Write it so another authorized participant could understand its meaning without the entire original conversation. A short, specific statement is usually more useful than an unstructured transcript. If a fact is uncertain, preserve that uncertainty instead of converting it into an authoritative-sounding instruction.

Choose the narrowest suitable scope

The Memory Notes interface includes Global, Organization, Project, Folder, and Conversation scopes, with availability governed by role and policy. The broadest choice is not automatically the best one. A formatting requirement for a single deliverable should not silently become guidance for every project. A project-specific term should remain associated with the work that gives it meaning. Consider who and what work should be affected. Use a conversation scope for a local requirement, a project or folder scope for related activity, and broader scopes only when the guidance genuinely belongs there and you are permitted to manage it. Ask Friday to help reason about the choice, then confirm the selected scope in the interface before saving.

Separate memory from sources and prompt layers

A Memory Note is not a substitute for an original document. A saved preference can shape an answer, but it does not prove the model read an attachment or can access a source. Similarly, a prompt layer describes how to approach work but does not grant a Tool permission. Keeping these distinctions clear prevents the application’s context mechanisms from being mistaken for one universal store of knowledge and authority. For a source-dependent claim, keep a reference to the relevant source and review the passage when needed. For a recurring behavioral preference, a concise note may be appropriate. For instructions that apply only to the present task, use the conversation. Choosing deliberately makes it easier to explain why information influenced an answer and where a correction should be made.

Verify the saved record

After using Add, inspect the note in the list and check its content, category, and scope. If you edited an existing note, confirm that the updated wording is visible. The application supports versioned changes where permitted, so maintaining a note does not require treating deletion and replacement as the default workflow. Do not rely solely on Friday saying “I will remember that.” A conversational promise is not a database record. If you asked for help drafting a note, distinguish the draft from the completed save. You can use Friday to improve clarity or suggest a scope, then carry out and verify the permitted interface action. That sequence keeps guidance useful while making persistence observable.

Review guidance as work changes

A note that was useful at the start of a project may become inaccurate after a decision changes. Review the relevant notes when a workflow, audience, or convention changes. Look for contradictions, outdated instructions, and details that belong in a narrower scope. Update the actual note where permitted rather than repeatedly correcting the assistant in unrelated conversations. Keep edits focused. State the current rule and the conditions under which it applies. Avoid embedding sensitive credentials or an entire private source document in a memory field. If the note needs provenance, include a concise reference that the intended audience can understand. The goal is maintainable context, not an ever-growing accumulation of assertions that nobody has checked.

Understand what memory does not promise

Saving a note is different from training the underlying model. It also does not guarantee that every future task, Agent, or conversation receives that note. Applicable context depends on scope, policy, and the runtime’s preparation process. If a response seems inconsistent with expected guidance, inspect the selected context and the note before concluding that it was lost. Ask Friday to explain which guidance is relevant where evidence is available, but do not ask it to invent an account of hidden state. If the interface and response disagree, preserve a small example for the operator. Good memory practice combines deliberate scope, concise wording, verified saves, and periodic review. It supports continuity without replacing source evidence, application permissions, or the user’s responsibility to evaluate important results.

Preserve decisions as inspectable records

The project publications describe continuity as a working discipline. At a useful stopping point, ask: “Prepare a handoff containing the current goal, approved constraints, source versions, accepted decisions, unfinished work, and the next action. Flag anything you cannot verify.” Review it before saving. Keep suggestions separate from decisions you have actually accepted. In the next session, provide the accepted handoff and ask the assistant to identify changes or conflicting sources. Confirm where the record is stored and that it can be reopened; a conversational statement that something was remembered is not sufficient evidence of persistence.

Explore the publications and annotated reading guide.