VAAS-X Cognitive Memory
Memori episodik persisten yang berlandaskan hasil untuk agen AI. Pengambilan data dengan CPU 14ms
Server MCP Terhosting
npx add-mcp 'https://mcp.vaasx.com/mcp'Terpasang ke Claude Code, Codex, Cursor, dan lainnya
Dokumentasi
Give your AI assistant a memory
Add one connector address to Claude, Cursor, VS Code or any agent that supports MCP. No code and no key to paste. Free to start, no card. Then see and manage everything it remembers in the app.
https://mcp.vaasx.com/mcp
Give your code and machines a memory
Install the SDK, connect a data source and start recalling what worked. On-device deployments run on the hardware you already own, with no GPU and no per-query bill. Free API key, no credit card. First query in under ten minutes →
pip install vaas-x

The app: ask a memory in plain words and the closest notes come back best first, each marked with how it turned out. These are sample notes.
Four ways in
One memory underneath, one account, one place to manage it. Start with the way that matches what you are doing.
THE APP
See and manage everything
One place for what your agents remember, what your SDK code writes, your devices, alerts and team. Sign in with your email; there is no password.
AGENTS (MCP)
Give any agent a memory
Add one connector address to any agent that supports MCP: Claude, Cursor, VS Code and others. No code, no key to paste. No more starting every session with an agent that has never met you.
SDK
Give your code a memory
Install vaas-x, then ingest, query and record what worked from Python. Free key, no card, first query in minutes.
BOOTSTRAP AND DEVICES
Point a data stream at it
Zero-config onboarding for sensors, machines and robots, a memory per device, and fleet search across devices with HiveMind.
Looking for the detail? Documentation · Developer guide · Reproduction guides · Whitepapers
What VAAS-X Actually Does
Most AI systems are stateless — they process data and discard it. VAAS-X gives any system a permanent, searchable record of everything it has experienced, retrievable in milliseconds, on the hardware already in place.
One infrastructure cost. Not a per-query bill that scales with your usage.
Cloud AI memory charges per prompt, per query, per token. As your system scales, so does your bill. VAAS-X runs as infrastructure — a fixed deployment cost regardless of how many queries your systems make or how many episodes they accumulate. Your costs are predictable from day one.
FILES OR A MEMORY SERVICE
When files stop being enough
If one agent on one machine is all you have, a markdown file is a fine place to start. Files stop being enough at three moments.
- Several agents need the same memory. A file sits in one folder on one machine. Connect Claude, Cursor and VS Code to the same VAAS-X account and each can recall from the same memory. Whether an agent reaches for it without being asked depends on the agent.
- A fact changes mid-project. A file keeps the old line next to the new one. In VAAS-X you mark the old note as superseded, so recall stops ranking it highly.
- You need what worked, not just what matches. A file gives back lines that look similar. VAAS-X records how each episode turned out and ranks recall toward what worked.
You do not have to take our word for the results: every benchmark has a reproduction guide you can run yourself. VAAS-X is not a chatbot and it is not a place to keep files.
RETRIEVAL ENGINE
Fast Memory Search — No GPU Required
Runs on the hardware you already own
VAAS-X retrieves the closest matching memory from over a million vectors (ANN-Benchmarks GloVe-100) in under 14ms — on a standard CPU server, with no GPU and no specialist hardware.
- 14ms on a standard EC2 CPU instance across 1.18M vectors (ANN-Benchmarks GloVe-100)
- 1.11ms on a desktop CPU — 297× faster than brute-force search on the same machine
- Runs on everything from a £30 microcontroller to an enterprise cluster
MEMORY FORMAT
Return to a task and your agent remembers what worked
Not just what happened — what worked
Each memory episode records what the system observed, what action it took, and what the outcome was. When something similar happens again, VAAS-X surfaces what worked before — not just the nearest match.
- Works across any domain without reconfiguration
- Prioritises episodes with successful outcomes
- Full audit trail on every stored memory
DATA SECURITY
Encryption-First Architecture
Built for sensitive operational data
Encrypted-data operation is on our roadmap for custom Enterprise deployments, so the infrastructure provider doesn't need plaintext access to what's stored. Contact us to confirm current availability for your compliance requirement.
- Encryption in transit (TLS 1.3) as standard on every tier
- Encrypted-data operation available for custom Enterprise deployments
- Relevant for healthcare, defence, and regulated industries
ILLUSTRATION
What “memory that survives a power cycle” looks like
One device, one memory store, three moments. Nothing the device recorded is lost when the power goes.
1. Running
The device records what it saw, what it did and how it turned out. Each of those is one episode.
2. Power off
Power is cut or the device resets. The device’s memory store is not touched.
3. Power on
The device asks what worked in a similar situation and gets the stored episodes back, best outcomes first.
The device’s memory store: every episode stays here the whole time.

The live demo, captured on 3 November 2026: one query that names no domain returns matches from three. Run it yourself.
ZERO-CONFIG ONBOARDING
Connect a Device. Memory Starts Immediately.
No integration project. No schema design.
Point any sensor array or data stream at Bootstrap and it begins building operational memory from the first datapoint — without knowing anything about the domain, the device, or the data format in advance.
- Validated: presented with 24 unlabelled NASA turbofan sensor channels — identified all 24 correctly with zero configuration
- Provisions a dedicated memory store per device automatically
- Operational in hours, not months
SCALE WITHOUT COST GROWTH
Add Devices. Costs Don't Follow.
Infrastructure pricing, not consumption pricing
Each new device or domain gets its own isolated memory store. Adding a hundred machines to your fleet doesn't multiply your AI costs — it extends the same substrate.
- Horizontal scaling by adding memory stores, not by upgrading infrastructure
- No migration or rebalancing required
- Encrypted and unencrypted deployments on the same architecture
EDGE & EMBODIED
Your fleet remembers every mission
No cloud connection required for on-device deployments
In on-device deployments VAAS-X runs entirely on the device. UAVs, robots, industrial controllers, and autonomous vehicles accumulate operational memory across their full lifetime — with no round-trip to the cloud and no dependency on connectivity.
- Validated on MicroPython hardware at ESP32 tier
- Identical codebase from microcontroller to enterprise cluster
- Memory persists across power cycles, missions, and resets
THE FULL SYSTEM
One Substrate. Every Domain. Unchanged Codebase.
Every capability above runs from the same infrastructure layer. Industrial sensors, autonomous platforms, AI agents, medical devices, and scientific instruments all use the same deployment — only the data differs. There is no domain-specific version of VAAS-X to integrate, maintain, or pay for separately.
- One codebase across every vertical
- No per-domain licensing or configuration
- Nine innovations unified into a single deployable layer
How It Works
No integration project. No specialist hardware. Memory starts from the first datapoint.
Step 1 — Connect
Point a data source at VAAS-X: a JSON address, an RSS or Atom feed, a CSV or JSON file, an MQTT topic or a WebSocket. Numbers become sensor channels, and text and ids are kept with their values so an agent can read them back. No schema design or configuration.
Step 2 — Accumulate
Every operational event is stored as a memory episode — what was observed, what happened, and what the outcome was. Memory builds continuously from the moment you connect.
Step 3 — Recall
When something happens, VAAS-X finds the closest matching episodes from history in milliseconds — on CPU with no GPU required, and with no cloud call in on-device deployments.
Step 4 — Improve
Because every episode records outcomes, the system surfaces what actually worked — not just what looked similar. Operational intelligence that compounds over time.
This is not a chatbot. This is not a vector database. This is operational memory infrastructure.
VAAS-X doesn't generate responses. It gives your systems the ability to learn from their own history — persistently, privately, and without ongoing AI compute costs.
Built for Sensitive Data
Patient records, classified operational data, and financial transactions stay encrypted in transit. Encrypted-data search is on our roadmap for custom Enterprise deployments — contact us to confirm current availability.
Millisecond Retrieval — No GPU Needed
14ms on a standard CPU server across 1.18 million vectors (ANN-Benchmarks GloVe-100). 1ms on desktop hardware. Retrieval that can run entirely on infrastructure you already own.
Connect Any Device in Minutes — Not Months
Bootstrap removes the integration project entirely. No schema design, no labelling, no domain expertise required. Validated on NASA's turbofan benchmark: 24 unlabelled sensor channels, 100% correct identification, zero configuration.
Memory That Prioritises What Worked
Standard search returns the nearest match. VAAS-X returns the nearest match with the best outcome — because every episode records what happened next. That distinction matters when the decision has consequences.
Validated Performance
April 2026. AWS EC2 m7i-flex.large. CPU-only. ANN-Benchmarks GloVe-100-Angular. 1,183,514 vectors. No GPU. No AVX-512. Mean and P95 are query latency; Recall@10 is the share of the true ten nearest neighbours returned. “vs Baseline” is measured against the baseline run on that row’s own machine (brute-force search on the desktop), so compare speed-ups within a row, not between rows.
| Configuration | Mean | P95 | Recall@10 | vs Baseline |
|---|---|---|---|---|
| Speed mode | 3.65ms | 4.94ms | 76.45% | 19x |
| Balanced mode | 9.32ms | 11.15ms | 89.05% | 7.4x |
| Quality mode | 13.99ms | 16.14ms | 93.15% | 4.9x |
| Desktop (i5-10600KF) | 1.11ms | 1.45ms | 76.5% | 297x |
1.11ms
Desktop mean
13.99ms
EC2 quality mean
297x
Desktop speedup
93.15%
EC2 Recall@10
100%
CMAPSS blind trial
9
Proprietary innovations
All benchmarks April 2026, AWS EC2 m7i-flex.large, CPU-only.
One Substrate. Deployed Across Domains.
The same codebase runs in every environment below. There is no domain-specific version to buy, integrate, or maintain. You deploy VAAS-X once — it operates wherever you point it.
INDUSTRIAL
Predictive Maintenance
Most industrial monitoring tells you when something has already failed. VAAS-X gives your machines the ability to recall every operational state that preceded a fault — across the entire fleet, without a per-site integration project.
- Bootstrap connects to any sensor array with no schema design or labelling
- 18.1 RMSE remaining-useful-life prediction on NASA CMAPSS FD001 — in line with standard numeric ML baselines on the same data, no feature engineering
- 100% zero-config sensor-channel identification on the same dataset, no prior schema
- No cloud dependency — runs on your existing on-premise hardware
EDGE & AUTONOMOUS SYSTEMS
Persistent Mission Memory
UAVs, ground robots, and autonomous vehicles generate valuable operational data on every mission — and discard it when they shut down. VAAS-X runs entirely on-device, accumulating persistent memory across the full platform lifetime with no cloud connection required.
- Runs on ESP32-class hardware — validated in live deployment
- Identical codebase from microcontroller to enterprise cluster
- Memory survives power cycles, disconnections, and resets
- No round-trip latency — recall happens on-device in milliseconds
AI INFRASTRUCTURE
Agent Memory That Persists
Every AI agent framework has the same gap: memory ends when the session ends. VAAS-X fills this at the infrastructure level — agents recall what happened in previous sessions, what decisions were made, and what outcomes followed.
- Cross-session memory — agents build on experience, not just prompts
- Outcome-weighted recall surfaces what actually worked
- Live deployment: 35,000+ episodes, autonomous loop operational
- No per-query cost growth as episode count scales
REGULATED INDUSTRIES
AI Over Sensitive Data — Without Compliance Risk
Healthcare, financial services, and defence need AI cognition over sensitive operational data — but cannot expose that data to an infrastructure provider. Encrypted-data operation, on our roadmap for custom Enterprise deployments, is designed to reduce the data processor relationship that creates compliance exposure.
- Encryption in transit (TLS 1.3) as standard on every tier
- Encrypted-data operation available for custom Enterprise deployments — contact us to confirm current availability
- GDPR, HIPAA, and sovereign air-gapped deployment pathways
SCIENTIFIC COMPUTING
Custom Computational Domains
The same substrate extends to advanced computational and scientific workloads for Enterprise customers. Details available under NDA.
- Same infrastructure, same API — no domain-specific build required
- Technical detail available under NDA
The substrate doesn't know which domain it's operating in. It doesn't need to. You provide the data stream — VAAS-X handles the rest.
Not a Better Vector Database. A Different Category.
Vector databases find semantically similar content. VAAS-X stores operational experience, recalls what produced the best outcomes, and can operate on-device, without cloud or GPU — at a fixed infrastructure cost that doesn't scale with query volume. These are not the same product.
See It Working on Your Data
Request a demo, discuss a pilot deployment, or ask a technical question. We can have memory running on your data stream faster than any integration project you've done before.