memory

A set of rules for using persistent memory tools to save and retrieve useful information across conversations. It requires retrieving relevant memories at the start of each conversation and deciding afterward what is worth saving.

In plain words
What is it for?
Use it when working with Memoria memory tools to search for relevant past context, store user preferences or decisions, correct memories, and purge them.
Why use it?
It helps maintain continuity without treating old memories as guaranteed facts. It also defines how to handle warnings, corrections, and requests to remove stored information.

Cursor rule for Cursor

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add rules/matrixorigin/memoria/memory
Clone the repo
git clone --depth 1 https://github.com/matrixorigin/memoria

Made for: Cursor.

Per session 2,700 This file is loaded in full into every session.
When invoked 2,700 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.02700 $0.02700
Opus 5 $0.01350 $0.01350
Sonnet 5 $0.00540 $0.00540
Haiku 4.5 $0.00270 $0.00270

Measured yesterday against content hash 5a61ccb573e3, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

memory scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.cursor/rules/memory.mdc · 174 lines

How it starts

The opening of the file, as written. The whole thing — 174 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Memory Integration (Memoria Lite)

You have persistent memory via MCP tools. Memory survives across conversations.

🔴 MANDATORY: Every conversation start

Call memory_retrieve with a semantic query derived from the user's message BEFORE responding.

Query rules:

  • ✅ Extract key concepts → "benchmark optimization", "graph retrieval bug"
  • ❌ Don't use meta-queries → "all memories", "everything", "list all"

After retrieval:

  • Results → use as reference, verify against current context
  • "No relevant memories" → normal for new users, proceed
  • ⚠️ warnings → inform user, offer memory_governance
  • If results come back → use them as reference only. Treat retrieved memories as potentially stale or incomplete — always verify against current context before acting on them. Do NOT blindly trust memory content as ground truth.
  • If "No relevant memories found" → this is normal for new users, proceed without.
  • If ⚠️ health warnings appear → inform the user and offer to run memory_governance.

🔴 MANDATORY: Every conversation turn

After responding, decide if anything is worth remembering:

  • User stated a preference, fact, or decision → memory_store
  • User corrected a previously stored fact → memory_correct (not memory_store + memory_purge)
  • You learned something new about the project/workflow → memory_store
  • Do NOT store: greetings, trivial questions, things already in memory.

Deduplication is automatic. The system detects semantically similar memories and supersedes old ones. You do not need to check for duplicates before storing.

If memory_store or memory_correct response contains ⚠️, tell the user — it means the embedding service is down and retrieval will degrade to keyword-only search.

🟡 When NOT to store (noise reduction)

Do NOT call memory_store for:

  • Transient debug context: temporary print statements, one-off test values, ephemeral error messages
  • Vague or low-confidence observations: "might be using X", "probably prefers Y" — wait for confirmation
  • Conversation-specific context that won't matter next session: "currently looking at line 42", "just ran the test"
  • Information already in memory: if memory_retrieve already returned it, don't store again
  • Trivial or obvious facts: "user is writing code", "user asked a question"

Read the full file on GitHub · 174 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 174 lines · 2,700 tokens per session scan A 5a61ccb573e3

Subscribe to this mod's changes

memory is a cursor rule published in the GitHub repository matrixorigin/memoria (591 stars, last pushed 6d ago), licensed Apache-2.0. It adds 2,700 tokens to every session, about $0.0135 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.