From 65de2993201290286a1a49efc33acbb0c8b0cf37 Mon Sep 17 00:00:00 2001
From: Luna <luna@example.com>
Date: Tue, 3 Mar 2026 20:34:17 +0100
Subject: [PATCH] added a memory cooldown

---
 .env.example  |  2 ++
 README.md     |  5 +++--
 src/config.js |  1 +
 src/memory.js | 24 +++++++++++++++++++++++-
 4 files changed, 29 insertions(+), 3 deletions(-)

diff --git a/.env.example b/.env.example
index 4db1856..f173303 100644
--- a/.env.example
+++ b/.env.example
@@ -13,4 +13,6 @@ ENABLE_DASHBOARD=false
 DASHBOARD_PORT=3000
 ENABLE_WEB_SEARCH=true
 OPENAI_API_KEY=your_openai_api_key
+# Memory retrieval cooldown (ms) before the same long-term entry can be reused
+MEMORY_COOLDOWN_MS=180000
 
diff --git a/README.md b/README.md
index 68bcb73..169cb54 100644
--- a/README.md
+++ b/README.md
@@ -102,8 +102,9 @@ Nova is a friendly, slightly witty Discord companion that chats naturally in DMs
 
    - **Embedding math:** `text-embedding-3-small` returns 1,536 floating-point numbers for each text chunk. That giant array is a vector map of the message’s meaning; similar moments land near each other in 1,536-dimensional space.
    - **What gets embedded:** After every user→bot turn, `recordInteraction()` (see [src/memory.js](src/memory.js)) bundles the pair, scores its importance, asks OpenAI for an embedding, and stores `{ content, embedding, importance, timestamp }` inside the SQLite tables.
-   - **Why so many numbers:** Cosine similarity needs raw vectors to compare new thoughts to past ones. When a fresh message arrives, `retrieveRelevantMemories()` embeds it too, calculates cosine similarity against every stored vector, adds a small importance boost, and returns the top five memories to inject into the system prompt.
-   - **Self-cleaning:** If the DB grows past the configured limits, low-importance items are trimmed, summaries compress the short-term transcript, and you can delete `data/memory.sqlite` to reset everything cleanly.
+- **Why so many numbers:** Cosine similarity needs raw vectors to compare new thoughts to past ones. When a fresh message arrives, `retrieveRelevantMemories()` embeds it too, calculates cosine similarity against every stored vector, adds a small importance boost, and returns the top five memories to inject into the system prompt.
+- **Memory cooldown:** `MEMORY_COOLDOWN_MS` (defaults to 180000 ms) keeps a long-term memory out of the retrieval window for a few minutes after it was just used so Nova has to pull fresh context before repeating herself, while still falling back automatically if there isn’t anything new to surface.
+- **Self-cleaning:** If the DB grows past the configured limits, low-importance items are trimmed, summaries compress the short-term transcript, and you can delete `data/memory.sqlite` to reset everything cleanly.
 
    ### Migrating legacy `memory.json`
    - Keep your original `data/memory.json` in place and delete/rename `data/memory.sqlite` before launching the bot.
diff --git a/src/config.js b/src/config.js
index 4cf4459..29ebc59 100644
--- a/src/config.js
+++ b/src/config.js
@@ -34,6 +34,7 @@ export const config = {
   legacyMemoryFile,
   summaryTriggerChars: 2200,
   memoryPruneThreshold: 0.2,
+  memoryCooldownMs: process.env.MEMORY_COOLDOWN_MS ? parseInt(process.env.MEMORY_COOLDOWN_MS, 10) : 3 * 60 * 1000,
   maxMemories: 8000,
   relevantMemoryCount: 3,
   longTermFetchLimit: 120,
diff --git a/src/memory.js b/src/memory.js
index 0fff446..1df3ae9 100644
--- a/src/memory.js
+++ b/src/memory.js
@@ -41,6 +41,15 @@ const parseEmbedding = (raw) => {
   }
 };
 
+const memoryUsageMap = new Map();
+
+const getMemoryUsageMapForUser = (userId) => {
+  if (!memoryUsageMap.has(userId)) {
+    memoryUsageMap.set(userId, new Map());
+  }
+  return memoryUsageMap.get(userId);
+};
+
 const __dirname = path.dirname(fileURLToPath(import.meta.url));
 const wasmDir = path.resolve(__dirname, '../node_modules/sql.js/dist');
 
@@ -333,8 +342,16 @@ const retrieveRelevantMemories = async (db, userId, query, options = {}) => {
   if (!rows.length) {
     return [];
   }
+  const now = Date.now();
+  const cooldown = config.memoryCooldownMs || 0;
+  const usage = memoryUsageMap.get(userId);
+  const eligibleRows =
+    cooldown && usage
+      ? rows.filter((entry) => now - (usage.get(entry.id) || 0) > cooldown)
+      : rows;
+  const rowsToScore = eligibleRows.length ? eligibleRows : rows;
   const queryEmbedding = await createEmbedding(query);
-  return rows
+  const scored = rowsToScore
     .map((entry) => {
       const embedding = parseEmbedding(entry.embedding);
       return {
@@ -345,6 +362,11 @@ const retrieveRelevantMemories = async (db, userId, query, options = {}) => {
     })
     .sort((a, b) => b.score - a.score)
     .slice(0, config.relevantMemoryCount);
+  if (scored.length) {
+    const usageMap = getMemoryUsageMapForUser(userId);
+    scored.forEach((entry) => usageMap.set(entry.id, now));
+  }
+  return scored;
 };
 
 export async function appendShortTerm(userId, role, content) {