Persistent Knowledge Graph Memory for AI Coding Agents. Adds long-term context (goals, strategies, preferences) to Cursor, Windsurf, & VS Code via MCP.
星數
16
工具
—
可用率
—
Neural Memory is a local-first personal cognitive memory assistant and second brain. It turns the desktop context you choose to share into an autonomous, searchable knowledge graph — anchoring real-world micro-feedback (like an "Ok" to an email quote), tracking commitments and decisions, and exposing your context directly to AI tools via the Model Context Protocol (MCP).
⚡ Quick Start: Choose Your Track
Neural Memory offers a Dual-Track Architecture designed for both casual MacBook users and developer power users:
Track B: Standalone Packed Installer (Recommended for Users)
Zero prerequisites. No Docker, no Python, no Homebrew, no terminal commands.
Track A: Open Local Development Stack (For Developers)
Full Docker Compose + Neo4j 5.26 Community Graph Database.
git clone https://github.com/Hexecu/mcp-neuralmemory.git
cd mcp-neuralmemory
./scripts/bootstrap.sh
Verifies installation: ./scripts/doctor.sh
Graph API: http://127.0.0.1:8765/health
Neo4j Browser: http://127.0.0.1:8774 (User: neo4j, Password in .env)
Stop without losing data: make down
Re-start: make up
🌟 Why Neural Memory is Different
🔒 PrivacyShield Redaction: In-memory regex and Luhn checksum filtering automatically masks credit cards, IBANs, API keys (sk-..., AIza..., GitHub tokens), and ignores password vaults (1Password, Bitwarden, Keychain).
🎯 Contextual Micro-Feedback Anchoring: Eliminates arbitrary length cutoffs. When you type "Ok, proceed with the 5,000 EUR quote", vision analysis anchors your consent directly to the sender, amount, and proposal document.
🌌 Temporal Knowledge Graph:
Regional Topic Clusters: Orbital regional gravity organizes entities into clean thematic galaxies without central collapse.
Time-Travel Scrubber: Interactive slider to replay memory history with Ebbinghaus memory decay ($R = e^{-\Delta t / S}$).
🤖 Agent-Ready Memory (MCP): Native Model Context Protocol server over stdio lets Claude Desktop, Cursor, Antigravity, and OpenCode recall past decisions, open commitments, and meeting notes.
🌙 Cognitive Dream Mode: Periodically prunes ephemeral low-level events (48h retention) and derives higher-order reflections and daily executive briefings.
The 5-stage massive test battery (make test-massive) and clean simulation (make verify-clean).
🤖 Connect to AI Agents (Claude Desktop & Cursor)
Neural Memory exposes your second brain directly to AI assistants:
# Auto-configure Claude Desktop:
./scripts/setup-mcp.sh --claude
Available MCP Tools for Agents:
recall_decisions(topic, person): Recalls past strategic agreements and approvals.
recall_commitments(status, debtor, creditor): Recalls deliverables, action items, and deadlines.
recall_meetings(topic, attendee): Recalls discussion points and participants from calls.
get_daily_briefing(date): Generates an executive summary of the day.
search(query): Deep hybrid search across your accumulated memory graph.
Example conversation:
User: "Claude, what did I agree to send to Marco this Friday?" Claude: "According to your Neural Memory, you committed to send the updated Cloud proposal slides to Marco Rossi by Friday at 18:00."
🛠️ Developer & Makefile Commands
make help # Show all available commands
make package # Build the standalone macOS DMG installer (Zero Docker required)
make verify-clean # Verify the standalone app in a clean, isolated MacBook simulation
make test-massive # Run the full 5-stage massive stress, scale, chaos & UI test suite
make test # Run Python unit & chaos tests (74 tests)
make e2e # Run end-to-end API integration tests (9 tests)
make doctor # Verify Docker, API, and Neo4j health
make lint # Run ruff linter
Persistent Knowledge Graph Memory for AI Coding Agents. Adds long-term context (goals, strategies, preferences) to Cursor, Windsurf, & VS Code via MCP.
上次檢查時間:2026-10-28。 Persistent Knowledge Graph Memory for AI Coding Agents. Adds long-term context (goals, strategies, preferences) to Cursor, Windsurf, & VS Code via MCP.