TapirusDB Edge
Downloads
Sovereign Embedded Intelligence

Zero Cloud. Zero Daemons.
Real-Time AI in Physical Devices.

Autonomous Robots, Smart Home Gateways, and On-Device Mobile Apps cannot afford 200ms cloud latencies, 1GB RAM footprints, or fatal network disconnects. TapirusDB embeds Relational SQL, HNSW Vectors, Knowledge Graphs, and JSON Documents directly into device silicon inside a single encrypted .tapir file.

Explore Use Cases Get Pre-compiled Binaries
< 4 MB
Idle RAM Footprint
0.51 µs
In-Process Vector-Graph
100%
Air-Gapped Offline
0 Bugs
Safe Rust Memory Invariant
The Paradigm Shift

How TapirusDB Replaces the Fragile Frankenstack

Until now, engineering intelligent edge devices required stitching together 4 to 5 incompatible libraries and cloud APIs. See the architectural contrast:

The Legacy Polyglot Stack (Before)
  • SQLite (Relational Only): Built in 2000 for simple tables. Incapable of semantic vector retrieval or native graph relationship traversals.
  • Cloud Vector DB Dependency (Pinecone/Qdrant): Devices require permanent WiFi/LTE connectivity. Introduces 100–300ms roundtrip network latency and massive cloud subscription costs.
  • C/C++ Segfault Hazards: Native ROS bag files, RocksDB, and custom FAISS builds are vulnerable to memory leaks and dangling pointer crashes.
  • Rapid Flash Wear-Out: Uncompressed high-frequency sensor writes burn out MicroSD cards and eMMC chips on Raspberry Pi in 6–12 months.
  • Heavy RAM Overhead: Running separate daemons consumes >500MB to 1.5GB RAM—impossible on constrained microcontrollers.
The TapirusDB In-Device Standard (Now)
  • Quad-Model in 1 Atomic File: Tables (SQL), Vectors (HNSW), Graphs (openCypher), and Documents (JSON) co-located in a single .tapir file.
  • 100% Offline & Zero Latency: Executes directly across in-process CPU memory registers (0.51 µs median latency) with zero network dependency.
  • Safe Rust Reliability: Enforced with #![forbid(unsafe_code)] in core modules. Zero buffer overflows, zero dangling pointers, and crash-proof WAL recovery.
  • Flash Wear Preservation: Built-in transparent LZ4 page compression reduces write amplification by 50%–70%, doubling hardware lifespan.
  • Ultra-Compact Footprint: Idle RAM under 4 MB, stripped binary under 1 MB, and initial file size of exactly 4,096 bytes.
Target Deployments

Three Core Physical AI Verticals

How TapirusDB powers next-generation robotics, smart home automation, and on-device private assistant apps.

Autonomous Robotics & AGVs

Spatial SLAM & Zero-Crash Robotics

Autonomous mobile robots (AMRs), warehouse automated guided vehicles (AGVs), and delivery drones operate under sub-millisecond physical deadlines. Cloud roundtrips cause collisions; C++ pointer bugs cause system freezes.

Topological Graphs Visual Loop Closure Power-Loss Immunity ROS / ROS2 C-ABI

Implementation Pattern: Camera keyframes are encoded into 384D vectors for loop-closure verification, while the openCypher graph maintains room connectivity for high-level path planning.

Smart Home & Industrial IoT

Wear-Proof Offline IoT Gateways

Smart hubs (Home Assistant, industrial PLCs, smart meter hubs) constantly write telemetry to low-cost MicroSD and eMMC storage. Traditional databases destroy flash storage and leak voice audio to the cloud.

LZ4 Flash Wear Reduction Offline Voice Intent Matching Zigbee/Matter Graphs Raspberry Pi Friendly

Implementation Pattern: Spoken commands are converted to text embeddings locally. TapirusDB matches semantic intent against home automation triggers without sending audio packets over the internet.

On-Device Mobile AI

Private Long-Term Memory for SLMs

Small Language Models (Gemma 2B, Phi-3, Llama 3.2 1B) run natively on modern Apple Neural Engines and Snapdragon NPUs, but suffer from complete amnesia when the user exits the app.

Episodic AI Memory Semantic Document Search ChaCha20-Poly1305 Zero Cloud Egress

Implementation Pattern: User facts and preferences are saved to personal knowledge graphs. Every turn queries memory in 0.51 µs, allowing models to remember family details and past context permanently.

Unified In-Process Kernel: 1 Single .tapir Container
Zero Inter-Process Serialization • Zero Cloud Middleware
┌────────────────────────────────────────────────────────────────────────────────────────┐ │ EDGE DEVICE SILICON / APPLICATION HOST │ │ (Raspberry Pi • NVIDIA Jetson • Apple Silicon • Qualcomm NPU) │ │ │ │ ┌────────────────────────────────────────────────────────────────────────────────┐ │ │ │ TapirusDB In-Process Engine (< 4 MB RAM) │ │ │ │ 100% Safe Rust • Direct Register Dereferencing │ │ │ └───────┬────────────────────┬───────────────────────┬───────────────────┬───────┘ │ │ │ │ │ │ │ │ ┌───────▼────────┐ ┌───────▼────────┐ ┌───────▼────────┐ ┌───────▼────────┐ │ │ │ 1. Relational │ │ 2. Schema-less │ │ 3. AI Vector │ │ 4. Knowledge │ │ │ │ SQL Tables │ │ JSON Document │ │ HNSW (SQ8) │ │ Graph Engine │ │ │ │ Telemetry Log │ │ Sensor Configs │ │ Camera/Voice │ │ Spatial Map │ │ │ └───────┬────────┘ └───────┬────────┘ └───────┬────────┘ └───────┬────────┘ │ │ └────────────────────┴───────────────────────┴───────────────────┘ │ │ │ Direct Memory Bus Access (0.51 µs) │ │ ▼ │ │ ┌──────────────────────────────────────────────┐ │ │ │ Page-Level ChaCha20-Poly1305 AEAD + LZ4 Comp │ │ │ └──────────────────────┬───────────────────────┘ │ │ │ Atomic WAL2 Frames (Power-Loss Safe) │ │ ▼ │ │ ┌──────────────────────────────────────────────┐ │ │ │ Single Encrypted Physical File: robot.tapir │ │ │ └──────────────────────────────────────────────┘ │ └────────────────────────────────────────────────────────────────────────────────────────┘
Hardware Readiness

Verified Across Modern Edge Silicon

TapirusDB is self-contained with no external system dependencies. Compiled binaries and C-ABI headers run natively across all major edge processors:

Hardware Platform Processor Architecture Primary Use-Case Idle RAM Verification Status
Raspberry Pi 4 / 5 ARM64 (Cortex-A72 / A76) Smart Home Hubs, Home Assistant, Field Data Loggers < 3.8 MB Verified Native
NVIDIA Jetson (Nano / Orin) ARM64 + CUDA Coprocessor Autonomous Mobile Robots (AMR), Visual Drones, Edge Vision < 4.0 MB Verified Native
Apple Silicon (A16–A18 / M1–M4) ARM64 (Apple Neural Engine) On-Device Private AI Assistants, Local-First iOS / macOS Apps < 3.5 MB Verified Native
Qualcomm Snapdragon 8 (Gen 2/3) ARM64 (Kryo / Hexagon NPU) Android Smartphone SLMs, Wearables, Automotive Cockpits < 4.2 MB Verified Native
Industrial Embedded x86 x86_64 (Intel Atom / Core, AMD Embedded) Factory Automation, CNC Controllers, SCADA Gateways < 3.9 MB Verified Native
WebAssembly / Edge V8 wasm32-unknown-unknown In-Browser Database Workbench, Cloudflare Workers < 4.5 MB Verified Native

Bring True Sovereignty to Your Edge Hardware

Stop paying monthly cloud vector bills and risking user privacy over network connections. Deploy a self-contained, crash-resilient intelligence engine in minutes.

Download CLI & SDKs View Edge Code Recipes Read Scientific Whitepaper