NEVERHALT ENGINE V2.0•AUTONOMOUS 24/7 AGENT CLOUD

Your agents never sleep.
So you can.

A persistent runtime that restarts crashes and catches silent stalls, for autonomous agents, trading bots, and FastMCP daemons. Crashes are easy. Silent stalls are expensive. NeverHalt catches both.

Prompt Cursor / Claude: "Set up NeverHalt hosting for me. Fetch https://neverhalt.app/llms.txt and package this project for 24/7 background execution with persistent storage on /app/data."

BUILT ON:

Ubuntu Linux 24.04•Docker Engine•NVMe Storage

deploy-target: /app/data (NVMe volume)

WATCHDOG ARMED

Drop your project archive (.zip or .tar.gz)

NeverHalt detects entrypoint, builds dependencies, and auto-mounts persistent NVMe storage to /app/data.

Max 50 MB•Python 3.11 / Node 20•Headless Chromium

UNIVERSAL RUNTIME COMPATIBILITY

If it runs on your machine, it runs on NeverHalt.

Pre-configured Linux environments with pre-baked dependencies. Click a workload to view runtime characteristics.

Fintech / HFT

Quant Trading (CCXT)

Sub-second arbitrage, orderbook polling, and websocket algorithmic trading bots. Never disconnects from Binance, Bybit or Kraken.

Auto-Entrypointrunner.py

Base Container512 MB - 1 GB

SAMPLE DEPLOYMENT SPECREADY

import ccxt
exchange = ccxt.binance({"enableRateLimit": True})
# Continuous 24/7 execution loop
while True:
    ticker = exchange.fetch_ticker("BTC/USDT")
    evaluate_arbitrage(ticker)

NVME STATE PRESERVATION

Keep your data safe.
Across reboots and crashes.

NeverHalt mounts an isolated NVMe block volume to /app/data. Your SQLite databases, trade ledgers, ML model weights, and session cookies persist permanently.

Automated daily rolling point-in-time snapshots

One-click .tar.gz backup downloads anytime

Zero data wiping when containers upgrade

trades_ledger.sqliteModified 2s ago • SQLite WAL mode

42.8 MB

agent_checkpoints.jsonModified 1m ago • State memory

1.2 MB

cookies_session.binModified 12m ago • Headless browser state

384 KB

NVMe Volume Allocation: 10 GB (44.4 MB used)ONLINE & MOUNTED

NEVERHALT ENGINE ARCHITECTURE•SUB-500MS AUTO-RECOVERY

Deploy without strain. Stay online forever.

Automated packaging and volume mounting on the left. Live supervisor crash recovery on the right.

Zero-Config Code Analysis

NeverHalt unpacks your project, detects entrypoint (bot.py, runner.py, scraper.py), and resolves dependencies with cached wheel layers.

Persistent Storage & RAM Allocation

Provisions dedicated container RAM and auto-mounts /app/data NVMe volume. SQLite, DuckDB, and state checkpoints survive forever.

Autonomous Launch & Watchdog Guard

Launches your daemon with live streaming logs. The supervisor detects unhandled crashes and re-boots the process in <500ms.

WATCHDOG SUPERVISOR PROOFEXIT_CODE_TRAP: SIGKILL / 137

SUPERVISOR WATCHDOG

INTERACTIVE DEMO

📊 Empirical Benchmark: p50 recovery is 219.7 ms (p95: 560.4 ms) measured over 50 SIGKILL cycles from signal dispatch to first process log.

btc-arbitrage-bot.pyPID: 4920

NVMe volume mounted at /app/data

running 24/7

Max tolerance: 500ms

Watchdog Telemetry Log:

18:42:01supervisor armed (watchdog: active)

18:42:02daemon process listening on /app/data

18:45:00zero memory leaks, 100% health

AGENTS RUNNING FOR

17d:04h:12m:00s

TRANSPARENT COMPETITIVE AUDIT

NeverHalt vs Fly.io, Railway, Render & a VPS

Crashes are easy to catch. Silent stalls are expensive. Here is an honest, dated breakdown of pricing and runtime capabilities across platforms.

Data verified on 24 October 2026 • Monthly pricing review schedule.

Scenario A: 1 Daemon Always-On (512MB RAM, 1GB NVMe)

Solo Bot / Trader

NeverHalt Starter

1 daemon, 512 MB RAM, NVMe /app/data volume, silent stall watchdog, 7-day trial (no card).

19 €

Fly.io Machines

shared-cpu-2x 512MB ($4.39) + $0.15/GB volume. Static outbound IP adds ~$3.60/mo.

~4.50 $

Render

Starter at $7/mo. Persistent disk billed additionally (requires paid plan).

~7 $ + disk

Railway

Hobby plan $5 base with $5 usage included, then metered per CPU/RAM.

~5 $ - 15 $

Self-Managed VPS (Coolify)

e.g. OVH VPS-1 ($6.49) or Hostinger KVM 2 ($8.99). All Linux, Docker & SSL setup on you.

~6 $ - 15 $

💡 Honest takeaway: For a single raw 512MB bot, unmanaged VPS or microVMs are cheaper. NeverHalt charges for zero DevOps, persistent volume setup, and silent stall detection.

Scenario B: 5 Daemons Always-On (2GB RAM each, 10GB NVMe)

High Density Fleet

NeverHalt Pro

Up to 5 active daemons, 2 GB RAM each, Chromium/Playwright pre-baked, snapshots, crash alerts.

39 €

Fly.io Machines

Official Fly.io estimate for 5 persistent 2GB always-on processes + volumes.

~71 $

Render / Railway

Metered resource accumulation across 10 GB total RAM and 5 continuous background containers.

~60 $ - 110 $

🎯 Density advantage: On multi-process workloads (10 GB aggregate RAM), NeverHalt Pro (39 €) cuts hosting costs by ~45% compared to Fly.io Machines (~$71).

Capabilities Matrix for Always-On Daemons

Scroll horizontally on mobile →

Feature	NeverHalt	Fly.io	Railway	VPS (Custom)
Deploy from ZIP in browser	✓ (Drag & Drop)	✗ (CLI only)	✗ (Git / CLI)	✗ (SSH manual)
Chromium & Playwright pre-baked	✓ (Kernel-ready)	✗ (User Dockerfile)	✗ (User Dockerfile)	✗ (Manual setup)
Persistent NVMe storage volume	✓ (/app/data)	✓ (Region-bound)	✓ (5 GB on Hobby)	✓ (Local disk)
Silent Stall Watchdog (Heartbeat & Log Silence)	✓ (Catches hung loops)	✗ (Crash-only restart)	✗ (Crash-only restart)	✗ (Manual scripts)
Circuit Breaker on crash loop	✓ (Halts after N crashes)	✗ (Infinite restart loop)	✗ (Infinite restart loop)	✗ (Crash loops indefinitely)
Clean shutdown (SIGTERM grace window)	✓ (Configurable grace_s)	✓ (Configurable)	✓ (SIGTERM 10s)	✓ (systemd TimeoutSec)
Isolation architecture	Docker cgroups/seccomp	Firecracker microVM (SOC 2)	Container runtime	Full KVM virtual machine

When NOT to choose NeverHalt

We believe in radical clarity. NeverHalt is purposely built for continuous daemons, bots, and MCP servers. It is not the right tool for everything:

•Multi-Region Clustering: If you need global geographic edge failover across 30 cities or auto-scaling clusters, use Fly.io.

•Managed Relational Databases & CI/CD: If you need managed Postgres, branching, and preview environments, use Railway or Render.

•Untrusted 3rd-Party Code & SOC 2: If you run arbitrary untrusted user code requiring hardware microVM virtualization or SOC 2 compliance today, use Fly Sprites or Daytona.

•DIY Linux Enthusiasts: If you already have a configured Linux VPS with systemd, Docker, and UFW and enjoy maintaining it yourself, you do not need NeverHalt.

Simple, Predictable Flat Pricing

No usage meter. No surprises. 7-day free trial on Starter limits (no credit card required).

Starter

Pro

BEST VALUE

For production AI agent swarms and quantitative trading bots.

39 €/mo

Up to 5 active daemons 24/7

2 GB dedicated RAM per process

Pre-installed Playwright & Chromium

FastMCP Streamable SSE & HTTP support

Instant Telegram/Discord crash alerts

Automated daily volume snapshots

Persistent high-speed NVMe storage

Customize

CUSTOM HARDWARE

For trading desks, quantitative funds and high-density agent swarms.

Custom

Unlimited active daemons 24/7

Configurable RAM (8 GB to 64 GB+)

Dedicated compute instance / bare-metal node

Dedicated static outbound IP address

Sub-minute precision crontabs

Private Telegram / Slack engineering war room

Custom Enterprise SLA agreement (Target 99.9%)

7-Day Free Trial: Terms, Expiry & Preservation Protocol

NO CREDIT CARD REQUIRED • ZERO AUTO-CHARGE

Strict 168-Hour Expiry

Your trial starts the second your daemon is deployed and runs continuously for 7 days (168h). At expiration, the process halts automatically to protect bare-metal compute.

14-Day NVMe Data Freeze

When the trial ends, your /app/data persistent storage (SQLite, tick history, bot memory) is frozen and preserved for 14 days, preventing any data loss.

Seamless Starter Upgrade

To prevent disruption to trading or monitoring, upgrade to the Starter Plan (19 €/mo) at any point during or after the trial to unlock permanent 24/7 unyielding uptime.

⚠️ Strict limits: 1 daemon, 512MB RAM, no crypto-mining or port-flooding (instant quarantine).

Need custom GPU compute or dedicated private hardware? .

SPECIFICATIONS & FREQUENTLY ASKED QUESTIONS

Everything you need to know.

Technical architecture, volume isolation, and watchdog mechanics.

You get full access to the Starter plan for exactly 7 days (168 hours) with zero credit card required. Deploy 1 continuous 24/7 daemon with 512 MB dedicated RAM, NVMe persistent volume, and the intelligent watchdog. When the 7 days end, your process pauses gracefully; your data on /app/data remains safe until you upgrade to Starter (€19/mo).

ZERO OVERAGE ANXIETY