Copilot Memory

Copilot is great, but it forgets everything the moment you close VS Code. This extension gives it a memory that actually sticks — across sessions, across days, across projects.

What it does

When you work with Copilot, important decisions get made. Which patterns to follow, which bugs to watch out for, why the code is structured the way it is. Without something to hold onto those facts, you end up re-explaining the same context every time you start a new session.

Copilot Memory stores that context locally and loads it back automatically every time VS Code starts. Copilot picks up where you left off without you having to remind it.

How memory works

There are two ways Copilot interacts with your saved memories:

Passively — on startup, recent memories are written into .github/copilot-instructions.md. Copilot reads this file automatically, so every session starts with your project context already loaded.

Actively — Copilot can call recall_memory and save_memory tools mid-conversation. When you make a decision worth keeping, it saves it. When you ask about something from a previous session, it searches your memory and pulls up what's relevant.

Memories are scoped per project and stored in VS Code's local extension storage. Nothing gets committed to your repo unless you want it to.

The `@mem` participant

For full session continuity — not just facts, but actual conversation threads — use @mem in Copilot Chat instead of talking to Copilot directly.

@mem how should we structure the auth layer?
@mem remember we decided to use JWT with 15-minute expiry
@mem remember this    ← saves your previous message

When you reopen VS Code after a @mem session, you'll get a prompt to resume where you left off. The last conversation is replayed as context so Copilot can continue naturally.

Requirements

VS Code 1.90 or later
GitHub Copilot (active subscription)

Commands

Command	What it does
`Copilot Memory: Remember This`	Save a memory manually via a quick input
`Copilot Memory: Show Memories`	Browse, copy, or delete saved memories
`Copilot Memory: Clear Project Memories`	Wipe all memories for the current project
`Copilot Memory: Show Status`	Check whether semantic search is active
`Copilot Memory: Setup Ollama`	Run the guided Ollama setup for semantic search

Memory categories

When saving, you pick a category. This helps with filtering and auto-detection:

Category	What goes here
`decision`	Architectural choices, design tradeoffs
`pattern`	Conventions the codebase follows
`context`	Background, constraints, team agreements
`bug`	Known issues and workarounds

Semantic search (optional)

By default, search is keyword-based. If you want meaning-based search — where "how do we handle authentication?" matches a memory about "JWT token validation" — you can enable it with Ollama running locally.

Toggle it on in settings:

"copilotMemory.semanticSearch.enabled": true

The extension will walk you through installing Ollama and pulling the embedding model. Nothing gets sent to the cloud — it all runs on your machine.

The status bar shows what's active at a glance:

database Copilot Memory — keyword search
sparkle Copilot Memory — semantic search via Ollama
warning Copilot Memory — Ollama enabled but not reachable

Setting	Default	Description
`copilotMemory.semanticSearch.enabled`	`false`	Enable Ollama semantic search
`copilotMemory.semanticSearch.ollamaUrl`	`http://localhost:11434`	Ollama server URL
`copilotMemory.semanticSearch.ollamaModel`	`nomic-embed-text`	Embedding model

Development

git clone https://github.com/philippham/copilot-memory
cd copilot-memory
npm install

Open in VS Code and press F5 to launch an Extension Development Host.

npm test    # run the test suite

Publishing

npm install -g vsce
vsce package    # builds a .vsix
vsce publish    # pushes to the marketplace

Full publishing guide: code.visualstudio.com/api/working-with-extensions/publishing-extension

License

MIT

Copilot Memory 101

philippham