ProjectMind

Universal Local-First AI Project Intelligence & Persistent Context Engine
Gives your coding agent a persistent memory, offline code intelligence, and a local dashboard — with zero cloud accounts required.

---

What is ProjectMind?

ProjectMind is not a coding agent or chatbot. It is a local intelligence and persistent memory layer that plugs directly into your existing coding agents:

• 🤖 Google Antigravity
• 🤖 Anthropic Claude Code
• 🤖 Google Gemini CLI
• 🤖 OpenAI Codex
• 🤖 Cursor
• 🤖 GitHub Copilot (VS Code)
• 🤖 Hermes Agent
• 🤖 DeepSeek Harness
• 🤖 OpenCode

Once installed, your agent automatically remembers your project, retrieves the right context before every task, and syncs changes — all offline, all local.

---

System Architecture

---

3-Layer Memory Model

Layer	Name	What it stores
1	Core Memory	Project identity, tech stack, key constraints — permanent
2	Project Knowledge	SQLite + FTS5 searchable knowledge, code graph, symbols, routes, models, ADRs
3	Task Context	Temporary, token-budgeted retrieval state per task

---

Smart Auto Workflow

Once installed into your agent, ProjectMind runs fully automatically:

1. On first use → initializes project memory (7-stage bootstrap)
2. Before each task → retrieves scoped context
3. After each edit → syncs changed files incrementally

---

Installation

Requires Node.js ≥ 18

---

🤖 Let Your Agent Do It (Recommended)

Just paste this prompt into your AI coding agent — it will detect itself, install ProjectMind globally, and verify everything automatically:

Set up ProjectMind so it works automatically in ALL my projects.

Steps:
1. Run: npm install -g @rootx202/projectmind
2. Detect which coding agent you are (antigravity / claude-code / gemini-cli / cursor / codex / github-copilot / hermes-agent / deepseek-harness / opencode)
3. Run: projectmind plugin install <your-agent-name> --scope global --yes
4. Run: projectmind doctor
5. Tell me which config files were updated and confirm everything is working.

Global scope configures ProjectMind at the user level — it activates automatically in every project you open, with zero per-project setup.

---

Manual Install (Global — works in all projects)

Step 1: Install globally

npm install -g @rootx202/projectmind

Step 2: Install your agent integration globally

# Pick your agent:
projectmind plugin install claude-code      --scope global --yes
projectmind plugin install antigravity      --scope global --yes
projectmind plugin install gemini-cli       --scope global --yes
projectmind plugin install cursor           --scope global --yes
projectmind plugin install codex            --scope global --yes
projectmind plugin install github-copilot  --scope global --yes
projectmind plugin install hermes-agent    --scope global --yes
projectmind plugin install deepseek-harness --scope global --yes
projectmind plugin install opencode         --scope global --yes

Step 3: Verify

projectmind doctor

---

Manual Install (Per-Project)

If you prefer to install only for a specific project:

npx -y @rootx202/projectmind plugin install claude-code --yes

Or via MCP directly:

npx -y @rootx202/projectmind mcp

---

Agent Compatibility Matrix

Agent	Native Plugin	MCP	Skills	Hooks	Auto Mode
Google Antigravity	✅	✅	✅	✅	✅
Claude Code	✅	✅	✅	✅	✅
Gemini CLI	—	✅	✅	✅	✅
Cursor	—	✅	✅	✅	✅
OpenAI Codex	—	✅	✅	Limited	✅
GitHub Copilot	—	✅	✅	—	✅
Hermes Agent	✅	✅	✅	✅	✅
DeepSeek Harness	—	✅	✅	✅	✅
OpenCode	—	✅	✅	✅	✅

---

CLI Commands

After installation, run diagnostics or manage your project memory:

# Run diagnostics
projectmind doctor

# View current context
projectmind context

# List installed plugins
projectmind plugin list

# Update plugins
projectmind update

CLI Screenshots

Command	Preview
projectmind doctor
projectmind context
projectmind plugin list
projectmind update

---

Key Features

• 🧠 Persistent Memory — Project knowledge survives agent restarts and new sessions
• 🌳 AST Code Analysis — Tree-sitter/WASM for Python, Go, Rust, Java, C, C++, C# + Babel for JS/TS
• 📊 Local Dashboard — HTML dashboard at 127.0.0.1 with memory layers, code graph, symbols, search, ADRs, and estimated token savings — zero CDN or external services
• 🔒 100% Offline & Private — Zero mandatory cloud accounts, zero paid APIs
• 🚫 Secret Protection — Auto-excludes .env, .pem, .key, .ssh, tokens
• 🛡️ Path Traversal Protection — Symlink escape & prompt-injection sanitization
• ♻️ Incremental Sync — Only re-indexes changed files using SHA-256 fingerprints
• 🏗️ Greenfield + Existing Projects — Works on new and existing codebases

---

Local Dashboard

The built-in dashboard runs on 127.0.0.1 only (never exposed to the network):

projectmind dashboard

Shows: memory layers · code graph · symbols · HTTP routes · database models · ADRs · search · events · token savings

---

Privacy & Security

ProjectMind is designed to be strictly local-first:

• ✅ All indexing, retrieval, and writes work 100% offline
• ✅ SQLite database stored in <project>/.projectmind/projectmind.db only
• ✅ Network access only for explicit, verified update checks (HTTPS + allowlist)
• ✅ Secrets automatically excluded from indexing
• ✅ .projectmind/ added to .gitignore automatically

---

Repository Structure

ProjectMind/
├── apps/
│   ├── cli/              # Main CLI entry point
│   ├── setup/            # Agent detection & idempotent installer
│   └── diagnostics/      # Runtime, SQLite, MCP & adapter health checks
├── packages/
│   ├── core/             # Project root detection, stable Project ID
│   ├── storage/          # SQLite3 + FTS5, 3-layer memory schema
│   ├── shared/           # Types, security validators, token estimator
│   ├── retrieval/        # BM25 + feature-hash reranker
│   ├── knowledge/        # Knowledge record management
│   ├── context-engine/   # Task context retrieval
│   ├── mcp-server/       # MCP stdio transport + local dashboard
│   ├── plugin-sdk/       # Plugin lifecycle, update, verification
│   └── agent-sdk/        # Smart Auto Mode engine + hook dispatcher
├── adapters/             # Per-agent native integrations
├── plugins/              # Plugin manifests for each agent
├── skills/               # Canonical SKILL.md bundle
├── templates/            # Default config template
├── docs/                 # Architecture, diagrams, screenshots
└── tests/                # Full test suite

---

Changelog

See CHANGELOG.md for the full version history.

Latest: v2.0.0 — Unified @rootx202/projectmind npm package, native plugin support for all agents, single MCP entry point via npx, removal of shell installers in favor of npm-backed integration.

---

Contributing

See CONTRIBUTING.md and PLUGIN_DEVELOPMENT.md for development guides.

---

License

MIT © Rootx202