Smart Memory
Your AI assistant forgets everything between sessions. Smart Memory fixes that.
Smart Memory is a VS Code extension that gives GitHub Copilot a persistent, searchable memory. It remembers your preferences, past mistakes, project decisions, and coding style — so you never have to repeat yourself.
Why?
Every time you start a new Copilot Chat session, it forgets:
- "I prefer Tailwind over plain CSS"
- "We use pnpm, not npm"
- "Never use
var in this project"
- "The auth service crashed last week because of a race condition"
Smart Memory saves all of this automatically and feeds it back to Copilot in every future conversation.
Quick Start (2 minutes)
1. Install the Extension
Install from a .vsix file or the VS Code Marketplace.
2. Start Talking to @memory
Open Copilot Chat and type:
@memory I prefer TypeScript strict mode for all projects
That's it. The memory is saved. Next time Copilot helps you, it will remember.
3. Let It Learn Automatically
Smart Memory also saves context by itself:
- Your tech stack — detected from your workspace
- Past failures — learned from git commit messages like "fix:", "revert:"
- Coding patterns — picked up during conversations
What Can It Do?
| Feature |
How It Works |
| Remember preferences |
Say "I prefer X" and it's saved forever |
| Search memories |
Type a question and it finds relevant context |
| Detect mistakes |
Warns you when editing near code that failed before |
| Generate project docs |
Creates a project brief from everything it knows |
| Team sharing |
Share memories with your team via file or AWS S3 |
| Cloud sync |
Sync memories across machines via DynamoDB |
| AI insights |
Claude on Bedrock analyzes your patterns and suggests improvements |
How to Use It
Talk to @memory in Chat
@memory remember that we use PostgreSQL, not MySQL
@memory what do you know about our auth service?
@memory what mistakes have I made recently?
In Copilot agent mode, these tools are available automatically:
| Tool |
What It Does |
save_memory |
Saves a preference, fact, or decision |
recall_memory |
Searches for relevant memories |
list_memories |
Shows all saved memories |
delete_memory |
Removes a memory |
update_memory |
Edits an existing memory |
check_past_failures |
Checks if your change might repeat a past mistake |
Plus 5 more tools for tags, stats, relations, export, and import.
Click the brain icon in the activity bar to see:
- Memories — All your saved memories, grouped and searchable
- Actions — All features organized into clean groups
- Dashboard — Browse and manage memories visually
AWS Integration (Optional)
Smart Memory works 100% offline by default. But if you want superpowers, connect it to AWS:
| AWS Service |
What It Adds |
Free Tier |
| Bedrock (Titan Embed) |
Better vector search quality |
1M tokens/month free |
| Bedrock (Claude Haiku) |
AI-powered memory insights & reports |
Pay per use |
| DynamoDB |
Sync memories across all your machines |
25 GB free |
| S3 |
Share memories with your team via links |
5 GB free |
| CloudWatch |
Track usage metrics and analytics |
10 metrics free |
Enable AWS in 3 Steps
- Run command:
Smart Memory: Configure AWS Credentials
- Enter your AWS Access Key and Secret
- Enable features you want:
Smart Memory: Enable Bedrock — better embeddingsSmart Memory: Enable Cloud Sync — DynamoDB syncSmart Memory: Export to S3 — team sharing- Set
smart-memory.analytics.enabled to true — usage metrics
Security: Credentials are stored in VS Code's encrypted SecretStorage, never in plain text. A minimal IAM policy is provided.
All Commands
| Command |
What It Does |
| Refresh |
Reload the memory sidebar |
| Search Memories |
Find memories by keyword |
| Clear Outdated |
Remove old, superseded memories |
| Export / Import |
Save or load memories as JSON files |
| Generate Project Brief |
Create a Markdown summary of your project |
| Resolve Conflicts |
Fix contradictory memories |
| Compact (GC) |
Merge duplicate memories |
| Enable Bedrock |
Switch to AWS-powered embeddings |
| Configure AWS |
Set your AWS credentials |
| Re-Embed All |
Recompute search vectors |
| Enable/Disable Cloud Sync |
Toggle DynamoDB sync |
| Force Sync Now |
Sync immediately |
| Export/Import S3 |
Team sharing via S3 |
| Show Analytics |
Open the usage dashboard |
| Generate AI Insights |
Claude analyzes your memory patterns |
Settings
| Setting |
Default |
What It Does |
smart-memory.embeddings.provider |
"local" |
"local" (offline) or "bedrock" (AWS) |
smart-memory.aws.region |
"us-east-1" |
AWS region for all services |
smart-memory.cloudSync.enabled |
false |
Turn on DynamoDB sync |
smart-memory.cloudSync.tableName |
"SmartMemorySync" |
DynamoDB table name |
smart-memory.s3.bucket |
"" |
S3 bucket for team sharing |
smart-memory.analytics.enabled |
false |
Send metrics to CloudWatch |
smart-memory.github.enabled |
true |
Watch git commits for failures |
Privacy & Security
- Everything is local by default — stored in
~/.smart-memory/memories.json
- No telemetry — the extension never phones home
- AWS is opt-in — no network calls unless you explicitly enable a cloud feature
- Credentials encrypted — AWS keys stored in VS Code SecretStorage
- CloudWatch metrics (when enabled) only send counts, never memory content
Architecture
You ← → Copilot Chat (@memory)
↓
Smart Memory Engine
├── Vector Search (local or Bedrock)
├── Memory Store (~/.smart-memory/)
├── Auto-tagger (21 rules)
├── Tiered Recall (hot/warm/cold)
└── Graph Relations
↓
Optional AWS Services
├── Bedrock → embeddings + Claude insights
├── DynamoDB → cross-machine sync
├── S3 → team sharing
└── CloudWatch → usage analytics
License
MIT
