⚡ NanoRecall: 100% Private Desktop Memory & Screen Search

🤝 Community, Issues & Support

• 🐛 Found a bug, screen capture issue, or memory leak? Please Open an Issue — reports are tracked and resolved quickly!
• 💬 Questions, UI ideas, or integration requests? Join our GitHub Discussions.
• ⭐ Find NanoRecall useful? Give it a star on GitHub — it helps more privacy-conscious users discover the project!
• ☕ Support development (USDT TRC-20): TDVbEdnpgNgoAhNcn1EwXxWHiR3RKLit5P (Verify on TronScan)

A high-performance, open-source, 100% private alternative to Microsoft Windows Recall.
Engineered with pure C99/SIMD, native Win32 capture, and zero cloud telemetry. Search anything you saw on your screen by natural language in milliseconds.

📦 Multi-Platform Installation

Platform / Manager	Installation Command
PyPI (pip)	pip install nanorecall
PyPI (uv)	uv add nanorecall
Conda (Anaconda.org)	conda install -c m_n_nik nanorecall
Ubuntu / Debian (APT PPA)	curl -sS https://eminsk.github.io/ppa/setup.sh | sudo bash sudo apt install python3-nanorecall
Ubuntu / Debian (.deb)	sudo dpkg -i python3-nanorecall_0.1.3-1_all.deb

---

🧩 Universal Compatibility Matrix

Runtime / Implementation	Supported Versions	Execution Mode	Status
CPython (Standard)	3.8, 3.9, 3.10, 3.11, 3.12, 3.13, 3.14, 3.15, 3.16 (Alpha)	Standard bytecode + GIL	✅ Fully Supported
CPython (Free-Threaded)	3.13t, 3.14t, 3.15t	Multi-core No-GIL (PEP 703)	✅ Fully Supported
PyPy (JIT Accelerated)	3.8, 3.9, 3.10, 3.11, 3.12	High-speed JIT tracing	✅ Fully Supported
Operating Systems	Windows (7, 8, 10, 11), Linux, macOS (Intel & Apple Silicon)	x86_64, ARM64	✅ Fully Supported

---

⚡ The Problem with Microsoft Recall

Feature	Microsoft Windows Recall	NanoRecall (This Project)
Privacy & Cloud	Unencrypted storage, telemetry risks	100% Local & Encrypted (Zero bytes leave your PC)
Hardware Lock-in	Requires expensive Copilot+ PC ($1500+) with 40+ TOPS NPU	Runs on any standard Intel / AMD / ARM CPU
Vector Engine	Heavy proprietary runtime	NanoVector (Pure C99 + AVX2, <120KB footprint)
Search Latency	Variable (Cloud / NPU overhead)	0.28 ms (Sub-millisecond exact semantic search)
Password Protection	Records sensitive credentials and cards	Privacy Shield: Auto-ignores 1Password, Bitwarden, KeePass, Incognito
Storage Footprint	Tens of gigabytes of raw frames	Smart Frame Differencing: Suppresses static frames (30–60 KB/frame)
Code Footprint	Gigabytes of OS bloatware	<200 KB pure codebase, <25MB RAM idle, 0.1% CPU

---

🌟 Key Highlights

• 100% Private & Offline: All frame embeddings and OCR snippets are stored in a single, local .nvec binary file. No accounts, no API keys, no internet required.
• Powered by NanoVector: Directly uses NanoVector's C99 AVX2/NEON vector search engine for instant 90-microsecond semantic retrieval.
• Intelligent Perceptual Frame Differencing: Continuously monitors your display; if the screen hasn't changed by >1.5% (reading, idle, away from desk), capture is skipped to conserve disk space and battery.
• Bare-Metal FASM SIMD Engine: Hand-crafted AVX2+FMA and SSE2 assembly microkernels executing screen differencing in 38 ns (26M fps) and vector dot products in 11 ns (90M ops/sec).
• Built-in Privacy Shield: Automatically detects active foreground windows and terminates recording whenever password managers (1Password, Bitwarden, KeePass), private browsing windows (Incognito, InPrivate), or crypto wallets are active.
• Dual Interface: Instant lightning-fast terminal CLI (nanorecall search ...) and a sleek dark-mode local web dashboard (nanorecall ui).

---

⚡ Bare-Metal FASM SIMD Hardware Engine

NanoRecall features hand-crafted x86-64 (AVX2+FMA) and x86 32-bit (SSE2) microkernels authored directly in Flat Assembler (FASM):

Microkernel	ISA / Vector Width	Throughput / Latency	Purpose
nanorecall_frame_diff_u8	AVX2 (vpsadbw) / SSE2 (psadbw)	38 ns / frame (26.3M fps)	32x32 grayscale screen diffing for static frame suppression
nanorecall_frame_diff_f32	AVX2+FMA / SSE2	45 ns / frame	Float32 perceptual delta comparison
nanorecall_vector_dot	AVX2+FMA (256-bit unrolled)	11 ns / op (90.9M ops/sec)	Fast vector inner product (dim=128)
nanorecall_cosine_similarity	AVX2+FMA / SSE2	14 ns / op	Episodic memory vector similarity
nanorecall_hamming_dist_u64	Hardware popcnt	2 ns / hash	Bitwise perceptual image hash hamming distance
nanorecall_mask_rect_rgba	AVX2 (vpbroadcastd)	<10 ns / region	High-speed RGBA privacy bounding box screen redaction

# Verify native bare-metal hardware engine:
nanorecall info
# Or run standalone assembly benchmarks in Windows / Linux:
cd asm && build.bat

---

🏛️ Architecture

                     ┌────────────────────────────────┐
                     │     Windows Desktop Display    │
                     └───────────────┬────────────────┘
                                     │
                        (Perceptual Diffing / 0.1% CPU)
                                     │
                     ┌───────────────▼────────────────┐
                     │    Native Win32 Screen Capture │
                     │  (GDI / BitBlt / DirectMemory) │
                     └───────────────┬────────────────┘
                                     │
                     ┌───────────────▼────────────────┐
                     │    Offline Text Extraction     │
                     │  (Windows.Media.Ocr / Local)   │
                     └───────────────┬────────────────┘
                                     │
                                     ├───────────────────────────────┐
                                     │ Text + Embeddings             │ Compressed Frame
                                     ▼                               ▼
                     ┌────────────────────────────────┐  ┌───────────────────────┐
                     │       NanoVector Core          │  │   Local Image Store   │
                     │  (C99 AVX2 + .nvec persistence)│  │   (WebP / 30-60 KB)   │
                     └───────────────┬────────────────┘  └───────────┬───────────┘
                                     │                               │
                                     └───────────────┬───────────────┘
                                                     ▼
                                     ┌────────────────────────────────┐
                                     │     Search & Recall Engine     │
                                     │   (CLI + Dark-Mode Dashboard)  │
                                     └────────────────────────────────┘

---

🚀 Interactive Google Colab Demo

Test NanoRecall interactively in your browser with zero local installation:

The Interactive Colab Notebook demonstrates:

• Backend & SIMD Detection: Verifying NanoRecall and NanoVector AVX2/NEON.
• Privacy Shield: Live simulation blocking 1Password/Incognito windows and redacting API keys.
• Desktop Memory & Search: Ingesting sample desktop screens and querying via natural language in <0.3 ms.
• Single-File .nvec Persistence: Saving and restoring the entire screen memory index.
• Colab CPU Benchmark: Benchmarking search latency (15–300 µs) and QPS over 10,000 frames.

---

🚀 Quickstart

1. Installation

pip install nanorecall
# or with uv
uv add nanorecall

(Requires Python 3.9+ on Windows, Linux, or macOS)

2. Capture Your Screen

Capture and index your current desktop state:

nanorecall capture

Output:

📸 Capturing desktop screen...
🔍 Extracting offline OCR text...
🧠 Indexing into NanoVector (Window: Active Window)...
✅ Indexed frame 2026-09-11_142315 (342 words, 98.4% change)

3. Natural Language Search (CLI)

Search for anything you saw hours, days, or weeks ago:

nanorecall search "github pull request sqlfluff"

Output:

🔍 Search Query: 'github pull request sqlfluff' (1 results found in 0.28 ms)

#1 [98% Match] Google Chrome — sqlfluff pull request 8449 github
   🕒 Captured: 2026-09-11_142315
   📝 Text:     merged upstream/main to pull in CI fix and added StarRocks test cases...
   🖼️  File:     C:\Users\M_N_N\.nanorecall\frames\2026-09-11_142315.webp

4. Interactive Web Dashboard

Launch the local dark-mode dashboard with visual timeline scrubber:

nanorecall ui

Then open http://127.0.0.1:8765 in your browser.

• Interactive daily timeline scrubber (scrub your day hour-by-hour).
• Instant search bar with live autocomplete and semantic similarity badges.
• Click any memory card to inspect the full-resolution screenshot with OCR highlights.

5. Background Daemon

Run NanoRecall silently in the background:

nanorecall daemon --interval 3.0

6. Model Context Protocol (MCP) Server

NanoRecall features a native Model Context Protocol (MCP) Server over JSON-RPC 2.0 stdio, exposing private desktop memory and real-time screen capture to Claude Desktop, Cursor, Windsurf, and Google Antigravity.

Quickstart

Run directly via CLI:

nanorecall mcp
# or using the dedicated entry point:
nanorecall-mcp

Client Configuration

Add to your claude_desktop_config.json, .cursor/mcp.json, or Antigravity MCP settings:

{
  "mcpServers": {
    "nanorecall": {
      "command": "nanorecall-mcp"
    }
  }
}

Or zero-install with uvx:

{
  "mcpServers": {
    "nanorecall": {
      "command": "uvx",
      "args": ["nanorecall", "mcp"]
    }
  }
}

Available MCP Tools

Tool	Purpose	Annotations
recall_search	Semantic vector search across screen history, OCR text, and past app windows.	readOnlyHint: true
recall_capture_now	Capture and index current desktop screen in real-time with privacy shielding.	readOnlyHint: false
recall_remember	Manually persist custom text facts, notes, or preferences into episodic vector memory.	readOnlyHint: false
recall_get_stats	Inspect total indexed frames, vector dimension, database size in KB, and storage path.	readOnlyHint: true

---

🛡️ Privacy Shield Rules

NanoRecall includes out-of-the-box protection rules. The following windows and sensitive tokens are automatically shielded from being recorded:

• Password Managers: 1Password, Bitwarden, KeePass, KeePassXC, LastPass, Dashlane, NordPass, Enpass.
• Private Browsing: Chrome Incognito, Edge InPrivate, Firefox Private Browsing, Tor Browser.
• Crypto Wallets: MetaMask, Phantom, Ledger Live, Trezor Suite, Exodus, Rabby.
• Credentials & Tokens: Credit card numbers, OpenAI API keys (sk-...), GitHub personal access tokens (ghp_...), and SSH/RSA private keys are automatically redacted.

---

📊 Telemetry & Benchmark

nanorecall stats

⚡ NanoRecall Engine Telemetry:
----------------------------------------
Total Indexed Frames: 1,420
Vector Dimension:     128D
Active Backend:       NanoVector (AVX2)
Local Database Size:  1.24 MB
Search Latency:       0.28 ms
Idle CPU Usage:       < 0.1%

---

🧪 Testing & Verification

Run the full pytest suite covering screen capture, memory persistence, OCR pipelines, CLI commands, and privacy redaction filters:

uv run --extra dev pytest -v
# or with standard pytest
pytest -v

All 13 tests pass with 100% success rate across Python 3.8 through 3.16 (including No-GIL free-threaded 3.13t–3.16t) and PyPy 3.8 through 3.12.

---

🌐 High-Performance Systems Ecosystem

nanorecall is developed by @eminsk as part of an open-source AI & systems engineering ecosystem:

• ⚡ NanoVector — Bare-metal C99/AVX2 vector search & episodic memory engine (~120KB) with Native MCP Server (pip install nanovector).
• 🧠 AgentJIT — Just-In-Time Compiler for AI Agent Trajectories with speculative de-optimization guards (pip install agentjit).
• ⚡ NanoGEMM — Bare-metal AVX2+FMA SIMD matrix multiplication engine in ~100KB for sub-microsecond CPU inference (pip install nanogemm).
• 🛒 avito-sdk — Headless Avito scraping & data extraction SDK with price drop tracking, Playwright cookies, Telegram/VK bots, and Native MCP Server (pip install avito-sdk).
• 📊 xlsx_vievers — Headless Excel formula engine (129+ functions), desktop spreadsheet viewer, SIMD SSE2 math, and Native MCP Server (pip install xlsx-viewer-pro).
• 📈 yfinance-ta-patterns — Candlestick & chart pattern scanner with AI Confluence Scoring, Backtesting, and Native MCP Server (pip install yfinance-ta-patterns).

---

🤝 Contributing

Contributions are warmly welcomed! Please submit issues or pull requests to improve OCR backends, UI features, or compression optimizations.

---

☕ Support, Community & Donations

If you find this project valuable and would like to support ongoing development:

• ⭐ Star the Repository: If NanoRecall helps you effortlessly search and recall your screen memory, give us a star on GitHub — it helps more developers discover private desktop memory!
• 💬 Join Discussions: Have ideas, use cases, or OCR models to suggest? Start or join a thread in GitHub Discussions!
• ☕ Donate (USDT TRC-20):
  TDVbEdnpgNgoAhNcn1EwXxWHiR3RKLit5P
  (Network: TRON / TRC-20 | Verify on TronScan)

---

📜 License

Distributed under the MIT License. Created by eminsk.