Memory Categories

PowerMemory doesn't store memories as one flat pile. Every memory belongs to a category and carries a priority. Together they decide what surfaces — and in what order — when you inject context into a new AI conversation.

This is the difference between "my AI remembers some stuff" and operational memory you can actually steer.

Priority tiers

Each memory is tagged P0–P3:

Priority	Meaning	Surfaces
P0	Critical — defines how your AI must behave	Always, first
P1	Important — active context you rely on	Almost always
P2	Useful — good to have when relevant	When space allows
P3	Background — nice but non-essential	Last, if room

When you inject memory into a fresh chat, P0 items lead. If the context budget is tight, P3 items are the first to drop.

The 12 categories

🎯 CALIBRATION

How your AI should communicate and collaborate with you.

• Use it for: tone, format, verbosity, language, feedback style.
• Avoid: one-off requests ("be brief just this once") — those aren't durable.
• Good: "Always give cold, direct feedback. No filler praise."
• Don't save: "Make this email shorter." (a task, not calibration)
• Priority: usually P0 — it shapes every interaction.

👤 IDENTITY

Who you are and the context you operate from.

• Use it for: your role, background, what you're building, your working context.
• Avoid: transient states ("I'm tired today").
• Good: "Solo founder running a SaaS + two service businesses."
• Don't save: "Currently on a call." (temporary)
• Priority: P0–P1.

🚫 HARD_CONSTRAINTS

Non-negotiable rules your AI must never violate.

• Use it for: absolute boundaries, safety rails, forbidden actions.
• Avoid: soft preferences (those go in PREFERENCE).
• Good: "Never modify production before verifying a backup exists."
• Don't save: "I prefer dark mode." (preference, not constraint)
• Priority: P0 — must always surface.

⚡ COGNITIVE_SHORTCUTS

Your decision heuristics and mental models.

• Use it for: how you reason, rules of thumb, default trade-offs.
• Avoid: one-time decisions (those go in DECISION).
• Good: "Prefer empirical verification over assumption before acting."
• Don't save: "Chose option B for this task." (a decision)
• Priority: P1.

📋 PROJECT

Active work and its current state across sessions.

• Use it for: what you're working on right now, where it stands.
• Avoid: finished projects with no ongoing relevance.
• Good: "Building v0.9 browser extension; OAuth flow done, UI pending."
• Don't save: "Thinking about maybe starting something." (not concrete)
• Priority: P1.

📘 LESSON

What you learned — so it's never re-learned.

• Use it for: what worked, what failed, what to never repeat.
• Avoid: restating obvious facts.
• Good: "A flag emitted is not a flag acted-on — test the downstream action too."
• Don't save: "Code can have bugs." (too generic)
• Priority: P1.

⚖️ DECISION

What you chose, why, and what was rejected.

• Use it for: concrete choices + the reasoning + the discarded alternative.
• Avoid: heuristics (those go in COGNITIVE_SHORTCUTS).
• Good: "Chose MkDocs over Docusaurus — lower maintenance for a solo founder."
• Don't save: "I usually prefer simple tools." (a heuristic)
• Priority: P1–P2.

💜 PREFERENCE

Your soft preferences — style, tools, defaults.

• Use it for: likes/dislikes that aren't hard rules.
• Avoid: absolute requirements (those go in HARD_CONSTRAINTS).
• Good: "Prefer concise responses with decision markers."
• Don't save: "Never send data to third parties." (a constraint)
• Priority: P2.

📌 FACT

Stable facts about you, your environment, your stack.

• Use it for: durable reference data that rarely changes.
• Avoid: fast-changing values.
• Good: "Primary domain is example.com; backend runs FastAPI."
• Don't save: "Server load is 40% right now." (volatile)
• Priority: P2.

💻 INFRA

Your servers, machines, networks, deployments.

• Use it for: hosts, IPs, deploy targets, hardware specs.
• Avoid: temporary connection details.
• Good: "Production VPS on Contabo; Caddy + FastAPI, static docs at /docs/."
• Don't save: "SSH session open right now." (transient)
• Priority: P2.

🔧 TOOL

Tools you use, their configs, and your patterns.

• Use it for: tooling setup, commands, working patterns.
• Avoid: one-off command runs.
• Good: "Deploy docs via scp to /opt/.../static/docs/ after strict build."
• Don't save: "Ran ls once." (not a durable pattern)
• Priority: P2.

💬 CROSS_AI_CHAT

Notable conversations with AI models worth preserving.

• Use it for: key handoffs, important cross-model exchanges, decisions reached with another AI.
• Avoid: routine chat noise.
• Good: "Cross-AI audit with ChatGPT flagged a legal wording risk on the landing page."
• Don't save: "Asked an AI a quick question." (no lasting value)
• Priority: P3.

What not to save (any category)

PowerMemory works best when the vault stays signal-rich. Skip:

• Temporary states ("on a call", "tired today")
• One-off tasks ("shorten this paragraph")
• Generic facts everyone knows
• Sensitive secrets (passwords, full keys, card numbers)

Advanced: how surfacing works

When you inject memory, PowerMemory ranks candidates by priority tier first, then by recency and relevance. A lightweight contradiction check flags when a new memory conflicts with an existing one (state update vs. genuine contradiction), so your vault stays consistent over time. Technical details live in the Privacy & Security Reference.