Choose a place by asking what you will need the information to do later. A standing preference should guide a new conversation. A task should remain visible until you finish it. A tracker should let you compare repeated observations.
You can ask a connected AI to propose the placement before saving. Approve the records you want, then check them in the dashboard.
A worked example: learning to sketch
Imagine someone learning to sketch for four weeks. Their conversation contains six different things:
| What they want to keep | Where it belongs | Why |
|---|---|---|
| “Give me one next step at a time.” | Core | A standing preference that can guide how the AI presents actions. |
| “Contour drawing helped me notice edges; rushing made the result less useful.” | Note | A specific observation worth finding when planning another exercise. |
| “Choose tomorrow's practice subject.” | Task | An action that still needs doing. |
| “Build a four-week sketching routine.” | Project | An ongoing objective with progress, resources, and next steps. |
| Minutes practiced each session | Tracker | Repeated observations with a consistent unit. |
| A checklist for reviewing a sketch | Skill | A reusable method for the AI to follow when requested. |
These are examples, not facts to add to your profile. Put your own information into this prompt:
Help me organize this information in Penny:
[paste my standing preferences, observations, actions, and ongoing goals]
Read relevant existing records first. Propose which items belong in Core,
Notes, Tasks, Projects, Trackers, or Skills, with a short reason for each.
Keep one-off details out of Core. Identify any records we can update
instead of duplicating. Show me the proposed contents before saving.
Core is loaded when the AI calls Penny's session-start tool; Notes need retrieval. A long explanation belongs in a note even if its short, durable rule belongs in Core. Read core memories and notes for that distinction.
After reviewing the proposal, copy this follow-up and name the records you approve:
Save the proposed records I approved: [name them here].
Use my private Penny memory and preserve unrelated existing content.
Create only those records. Link relevant records to the Project if we
approved one. Retrieve each saved result and list where I can inspect it.
Report any failed save separately from successful saves.
Verify the arrangement
Open Core, Notes, Tasks, or Projects for the approved records. Check the exact wording, task status, and Project links. Then inspect Trackers for its definition and Skills for its instructions.
For the sketching example, a tracker entry of 20 should mean 20 minutes in one session. A review skill should contain the checklist; saving it does not schedule a run. The Project can link both without copying their contents into Progress.
Use task instructions, tracker instructions, and skill instructions when you need the individual controls. If an approved record is missing, check the save result before repeating the request.