NeuroSync

Stop repeating project context to AI every session.

NeuroSync builds persistent, local project memory in the background and injects it into the files AI assistants already read.

Why teams install NeuroSync

• Automatic memory from real engineering signals: file changes, diagnostics fixes, terminal failures, and git history.
• Local-first encryption with zero required cloud dependency.
• Multi-agent context outputs for Copilot, Claude-style workflows, Cursor-style rules, and MCP clients.
• Token-efficient context generation with compression and noise filtering.

What you get in 60 seconds

1. Install extension.
2. Open your workspace.
3. Keep coding normally.
4. NeuroSync continuously updates context artifacts like CLAUDE.md and .github/copilot-instructions.md.

Features

Works automatically — zero setup

• Watches your files in the background as you code.
• Learns from file changes, error fixes, terminal failures, and git history.
• Continuously updates context files that AI assistants already read.
• Detects your project stack automatically.
• Automatically protects sensitive memory files from being committed to git.

Memory sidebar

NeuroSync adds a panel to the VS Code activity bar showing your project memory — what has been learned, categorized by type (fixes, conventions, decisions, gotchas), with timestamps.

Using the sidebar

Click the NeuroSync icon in the VS Code activity bar to open the memory panel.

Session summary

At the end of every session, NeuroSync writes a .agent-mem/last-session.md file summarizing what changed, which files were edited, and what was observed. Useful for picking up where you left off or briefing an AI agent on recent activity.

Ask AI about your project memory

When using GitHub Copilot agent mode, NeuroSync registers built-in tools that Copilot can call automatically mid-conversation:

• Query memory — searches what NeuroSync has learned for a topic
• Get rules — retrieves your project rules and lessons
• Save lesson — lets Copilot permanently save a correction so it never repeats the same mistake

These work without any extra setup.

Private and local

• All memory stays in your workspace — nothing leaves your machine.
• Encrypted local storage, no cloud account required.

Works with all major AI tools

• GitHub Copilot, Claude, Cursor, Windsurf, Antigravity, and any MCP-compatible client.
• Updates CLAUDE.md, AGENTS.md, .github/copilot-instructions.md, .cursorrules, .windsurfrules, and more.

How it works

Just install and open your workspace. NeuroSync runs silently in the background.

Context artifacts

Default outputs:

• CLAUDE.md
• AGENTS.md
• AGENT.md
• .github/copilot-instructions.md
• .agents/rules/neurosync-context.md
• .cursorrules
• .windsurfrules

Internal outputs:

• .neurosync/rules/neurosync_auto.md
• .neurosync/agent-rules.md
• .neurosync/ip_rules.md

Compatibility

NeuroSync detects your stack automatically from project files. No configuration needed.

If your project file exists (package.json, pubspec.yaml, go.mod, Cargo.toml, pom.xml, build.gradle, requirements.txt, *.csproj), NeuroSync picks it up automatically.

Installation

1. Install the extension.
2. Open any workspace.
3. NeuroSync activates and starts observation automatically.

No mandatory initial setup is required for core functionality.

Quickstart

1. Install and open your workspace — NeuroSync starts automatically.
2. Keep coding as normal. No setup needed.
3. Check CLAUDE.md or .github/copilot-instructions.md after a few minutes to see memory being built.

Commands

• NeuroSync: Save Note
• NeuroSync: View Memory
• NeuroSync: Search Memory
• NeuroSync: Sync Context Files

Settings

• neurosync.autoWatch
• neurosync.syncTargets
• neurosync.syncOnSave
• neurosync.compressionMode

Recommended baseline:

• Keep neurosync.autoWatch enabled.
• Keep neurosync.compressionMode at aggressive for maximum token efficiency (balanced for better readability with moderate savings, off for full verbose context).

Power user: teach AI your project rules (optional)

By default, NeuroSync learns from what you do. But you can also tell it things explicitly — and those will be injected into every AI agent's context automatically.

This is completely optional. The extension works without it.

Project rules — .neurosync/ip_rules.md

Write down the things you want every AI to always know and follow for this project.

Example:

# My Project
- Customer support dashboard for a SaaS product.
- Primary users: support agents and team leads.

# Rules
- Always use British English in UI text.
- Never suggest deleting user data — always soft delete.
- Payments handled by Stripe only — never suggest alternatives.

Save the file — AI agents pick it up within ~5 seconds. No need to repeat yourself every session.

For an instant update, run NeuroSync: Sync Context Files from the command palette (Ctrl+Shift+P).

Lessons log — .agent/lessons.md

A running list of things AI should avoid right now — bugs, known workarounds, or temporary constraints.

Example:

# Lessons
- [2026-05-13] Do not use the date picker for time ranges — it breaks on mobile. Use RangePicker instead.
- [2026-05-10] The export button has a known race condition. Do not refactor until the async fix is merged.

When the issue is fixed, delete the line. Agents stop avoiding it immediately.

Tip: Even 3–5 lines in ip_rules.md noticeably improves AI response quality. Start small.

MCP server

The MCP server is already compiled and bundled inside the installed extension. No build step needed.

To connect it to an MCP-compatible AI client (Claude, Cursor, Windsurf, etc.), create a .mcp.json file in your project root:

{
  "mcpServers": {
    "neurosync": {
      "command": "node",
      "args": ["/path/to/your/.vscode/extensions/touhidion.neurosync-<version>/out/mcp/stdioServer.js"],
      "env": {
        "NEUROSYNC_WORKSPACE": "${workspaceFolder}"
      }
    }
  }
}

Find your exact extension path: run code --list-extensions --show-versions and look for touhidion.neurosync.

The AI client launches the server automatically as a child process — you never start it manually.

If you cloned the repo (development only): npm run mcp:build compiles the server, npm run mcp:start runs it standalone for testing.

Storage layout

• .agent-mem/observations/YYYY-MM-DD.md — daily plain-text log of everything learned that day (auto and manual). Open any file directly in VS Code from the Explorer panel, or browse them via the View All button in the sidebar which shows the full list in a searchable panel. Readable by any AI tool or agent directly.
• .agent-mem/index.md — summary index of all observations grouped by date, with token estimates.
• .agent-mem/last-session.md — last session summary (files changed, observations created).
• .neurosync/.context-key — encryption key (never commit — auto-added to .gitignore).
• .neurosync/store.json — encrypted memory store (never commit — auto-added to .gitignore).

Troubleshooting

MCP server exits or appears idle

• Verify the server path in your .mcp.json points to the correct out/mcp/stdioServer.js inside the installed extension folder.
• The server is launched automatically by the AI client — do not start it manually.
• If run manually outside an MCP client, idle behavior (no output) is expected — it only responds to JSON-RPC messages on stdin.

Debug console noise from other extensions

• NeuroSync logs to its own Output channel.
• Open Output panel and select NeuroSync from the channel dropdown.

No observations appearing

• Confirm workspace folder is open in extension host.
• Confirm neurosync.autoWatch is enabled.
• Save a supported source or config file and wait for debounced sync.

Privacy

NeuroSync is local-first for memory storage and context generation.