Shodh-Memory: A Cognitive Memory System for Edge-Native AI Agents

Q: How is shodh-memory different from a vector database?

Vector databases give you similarity search. Shodh-memory gives you cognition — memories strengthen when accessed together (Hebbian learning), decay naturally over time (power-law forgetting), and form associative networks via a knowledge graph. It's the difference between storage and memory.

Q: Does shodh-memory require an internet connection?

No. Shodh-memory runs 100% offline. The embeddings, vector index, knowledge graph — everything runs locally. Perfect for edge devices, air-gapped systems, or anywhere you need data privacy.

Q: What's the memory overhead?

The binary is ~30MB. Models add ~50MB (22MB MiniLM embeddings + 14MB NER model + 14MB ONNX runtime). Each memory entry uses roughly 2-5KB. A system with 10,000 memories uses approximately 50MB of storage.

Q: Can shodh-memory run on a Raspberry Pi?

Yes. Shodh-memory is designed for edge deployment. It runs on Raspberry Pi Zero, Jetson Nano, industrial PCs, and other resource-constrained devices. Graph lookups are under 1 microsecond.

Q: How does memory decay work?

Shodh-memory uses a hybrid model: exponential decay for the first 3 days (consolidation phase), then power-law decay for long-term retention. Memories accessed 10+ times become potentiated and decay 10x slower. Based on Wixted & Ebbesen (1991).

Q: What is Hebbian learning in AI agent memory?

Cells that fire together, wire together. When memories are accessed together, their connection strengthens. When memories compete, interference effects occur. It's how biological brains work, now applied to AI agent memory.

Q: Is there a cloud version?

No, and that's intentional. Shodh-memory is built for local-first, privacy-preserving AI. Your agent's memories stay on your hardware. If you need multi-device sync, you can replicate the RocksDB storage yourself.

Q: What languages and frameworks does shodh-memory support?

The core is Rust. We provide: an MCP server (for Claude, Cursor, and other AI agents), Python bindings (via PyO3/maturin), and a REST API. The Rust crate can be embedded directly in your application.

Q: How do I contribute?

Check out github.com/varun29ankuS/shodh-memory. Open issues, submit PRs, or join discussions. The codebase is well-documented with 688+ tests. All constants have neuroscience citations.

Varun Sharma

doi:10.5281/zenodo.18668709

[edge]

Where the cloud can't follow

A robot arm can't wait 300ms for a round trip. A drone loses its uplink. A factory's process data can't leave the building. Every memory system that depends on a cloud API fails at exactly the moment robotics needs it most. shodh-memory lives on the machine — and learns there.

why-edge.md

# Why Edge Memory Matters

Cloud-based AI has three fatal flaws for robotics:

  1. LATENCY      - 100-500ms round trip kills real-time control
  2. CONNECTIVITY - Factories, farms, vehicles lose connection
  3. PRIVACY      - Operational data leaving premises = IP risk

Shodh-memory solves all three:

  + <1us graph lookup, 34-58ms semantic search
  + Single binary, zero network dependencies
  + All data stays on-device, always

30MB

single binary, no runtime deps

<1μs

graph lookup on-device

512MB

minimum RAM (Pi Zero territory)

0

network calls, ever

[01]

On the factory floor

Production environments measure downtime in thousands of dollars per minute. Memory here has to be deterministic, local, and faster than the control loop.

industrial-robotics.rs

🤖

Industrial Robotics

Memory for robot arms and autonomous manufacturing systems. Learn tool preferences, remember failure modes, adapt to new parts without retraining.

<50ms recall latencyOffline operationReal-time learning

A welding robot remembers which torch angles work best for different metal thicknesses.

smart-manufacturing.rs

🏭

Smart Manufacturing

Production line AI that learns from experience. Remember optimal machine settings, predict maintenance needs, adapt to product variations.

Air-gapped deploymentMulti-agent coordinationTemporal patterns

A CNC controller remembers that aluminum batch #4523 requires different feed rates.

warehouse-automation.rs

📦

Warehouse Automation

Memory for autonomous forklifts and picking robots. Learn warehouse layouts, remember item locations, optimize routes.

Real-time coordinationSpatial patternsPersistent across shifts

A picking robot learns that aisle 7 gets congested at 2pm.

[02]

In the field

Vehicles, farms, and contested airspace share one constraint: the network is a luxury. Memory has to survive disconnection — and restarts — without losing what it learned.

autonomous-vehicles.rs

🚗

Autonomous Vehicles

Edge-deployed memory for self-driving platforms and drones. Store learned routes, remember unusual traffic patterns, maintain situational awareness.

Sub-millisecond lookupNo cloud dependencyPersistent across restarts

An autonomous delivery drone remembers that Building 7 has a tricky landing zone.

agricultural-robotics.rs

🌾

Agricultural Robotics

Memory for farm robots and smart irrigation. Learn field conditions, remember crop-specific needs, adapt to seasonal changes.

Low-power edgeWeather-resilientMulti-season learning

A harvesting robot remembers soil conditions in different field sections.

defense-and-aerospace.rs

🛡️

Defense & Aerospace

Air-gapped memory for mission-critical systems. No external dependencies, no network calls, complete operational sovereignty.

Zero network requiredDeterministic recallSingle auditable binary

A reconnaissance drone maintains mission context across communication blackouts.

[03]

Close to people

When memory holds health data or family routines, the only acceptable place for it is the device itself. Nothing leaves. Nothing phones home.

medical-devices.rs

🏥

Medical Devices

Local-only memory for medical AI. Patient-specific preferences, treatment history, and usage patterns stored on the device — never in someone else's cloud.

Local-only storageDeterministic recallPersistent history

An insulin pump AI learns individual patient glucose patterns over time.

smart-home-hubs.rs

🏠

Smart Home Hubs

Privacy-first home automation memory. Learn family routines, remember preferences, adapt to seasonal patterns.

Local-only processingFamily-safe privacyLow power (<5W)

A home assistant learns that Tuesday evenings mean workout music.

[04]

What it runs on

If your hardware can run Linux, it can probably run shodh-memory.

Minimum

+ARM Cortex-A53
+512MB RAM
+100MB storage
+Linux/RTOS

Tested Platforms

+Raspberry Pi 4/5
+NVIDIA Jetson
+BeagleBone
+Industrial PCs

Your robots already work where the cloud doesn't.

Now their memory can too. One binary, one command.

$ npm install View Source Talk to us about a deployment