Scanning every GPU cloud on Earth, live

Stop renting one cloud. Start renting the cheapest one.

Run one command from your own terminal. Vivacious Cloud shops every GPU provider we're connected to in real time — not just the big four — deploys your job to whichever is cheapest right now, and jumps it to a better rate the moment one opens up. No dashboards, no DevOps, no babysitting.

  • Spot prices move hourly — you're paying yesterday's rate on today's GPU.
  • 🗂️ Comparing a dozen providers by hand costs you an afternoon per job.
  • 🔒 Lock into one provider, and you eat their price the second it moves against you.
  • 🧹 Nobody deletes the training data after — it just sits there, a bill with no purpose.
0 GPU providers in our network
0 rate re-check interval
0 idle cost when you're not training

Why founders switch

Built for one person who doesn't have a DevOps team.

Because you are that person.

🎯

Zero-config deployment

No YAML, no cluster config, no infrastructure-as-code. One terminal command in, finished weights out.

💸

Real-time price arbitrage

We're always watching what every provider charges right now — not what they charged when you last checked.

🔁

Mid-job provider switching

If your provider gets slow or expensive mid-run, we checkpoint, migrate, and resume — automatically.

🧊

Zero-egress storage

Your data lives in Cloudflare R2, not locked to one cloud — moving providers never costs a transfer fee.

Get started

Three steps. That's the whole onboarding.

  1. 01

    Make your account

    Sign up once on the site — this is the only time you'll need a browser.

  2. 02

    Add your account details

    Card on file, verified. After this, billing runs quietly in the background.

  3. 03

    Run the terminal commands

    Point us at your dataset and your script. Everything else happens off-screen.

From your machine

No dashboard required. Just your terminal.

Works the same in Windows, Mac, or Linux terminals. Copy, paste, run.

1 Install the CLI
Mac / Linux
curl -fsSL https://vivaciouscloud.com/install.sh | sh
Windows
iwr https://vivaciouscloud.com/install.ps1 -useb | iex
2 Log in from the CLI
terminal
vivacious login anirudha-s
3 Prepare your dataset locally
terminal
vivacious prepare <path>
Inspects files and computes SHA-256 manifest locally. No cloud transfer or GPU compute.
4 Permit the workload (Pre-Flight Check)
Step 4a (Full Fine-Tuning - Default)
vivacious permit anirudha-s ambition --model <model-id>
Runs full fine-tuning estimation by default. Models wholesale spot rate + 30% platform markup.
Step 4b (LoRA)
vivacious permit anirudha-s ambition --model <model-id> --method lora
LoRA parameter-efficient fine-tuning requiring less GPU VRAM.
Step 4c (QLoRA)
vivacious permit anirudha-s ambition --model <model-id> --method qlora
QLoRA quantized 4-bit fine-tuning for maximum cost savings.
5 Deploy the job
terminal
vivacious deploy anirudha-s
Uploads prepared dataset to Vivacious Storage and launches cheapest GPU instance across providers.
6 Monitor Balance & Exposure
terminal
vivacious balance
Displays server-authoritative live balance, total credit paid, and total spent.
7 Download Model Weights
terminal
vivacious download <job-id>
Downloads signed model weights tarball once job is completed and billing is settled.
Live Router Traffic Logs
router-traffic.log LIVE FEED
[16:14:00] Connecting to router feed...
[16:14:01] Established connection. Live sweep active.
1 Install the CLI
Mac / Linux
curl -fsSL https://vivaciouscloud.com/install.sh | sh
Windows
iwr https://vivaciouscloud.com/install.ps1 -useb | iex
2 Log in from the CLI
terminal
vivacious login anirudha-s
3 Prepare Dataset & Checkpoint Locally
Step 3a (Prepare New Dataset)
vivacious prepare dataset ./new_instructions.jsonl
Computes streaming SHA-256 fingerprint locally. Zero cloud transfer or GPU compute.
Step 3b (Stage Base Weights / Checkpoint)
vivacious prepare checkpoint ./my_custom_model.tar.gz
Scans safetensors allowlist & parses model config locally. Verifies zero pickle execution.
4 Permit the Retraining Workload
Full Fine-Tuning Retrain
vivacious permit anirudha-s ambition --model custom_model
Sizing engine sizes GPU requirements based on staged checkpoint architecture metadata.
LoRA Retrain on Custom Weights
vivacious permit anirudha-s ambition --model custom_model --method lora
Applies LoRA adapter fine-tuning directly on top of your existing custom base weights.
QLoRA (4-bit) Retrain
vivacious permit anirudha-s ambition --model custom_model --method qlora
Quantized 4-bit retraining for maximum memory efficiency and lowest spot rate.
5 Deploy the Retraining Job
terminal
vivacious deploy anirudha-s
Uploads both dataset and custom weights to Vivacious Storage and launches isolated GPU sandbox.
6 Monitor Telemetry & Balance
Track Live Job Progress
vivacious status <job-id>
Displays step progress, live training loss, VRAM consumption, and elapsed hours.
Check Balance & Spend
vivacious balance
Displays server-authoritative live balance, total credit paid, and total spent.
7 Download Retrained Weights
terminal
vivacious download <job-id>
Downloads signed model weights tarball once job is completed and billing is settled.
Live Router Traffic Logs
router-traffic.log LIVE FEED
[16:14:00] Connecting to router feed...
[16:14:01] Established connection. Live sweep active.

Billing

You pay a slice. Never a subscription.

No monthly plans, no seat fees. Vivacious Cloud takes a small percentage of the compute cost it finds for you — the cheaper your job runs, the less you pay us too.

>0%

Convenience margin

Added on top of the live wholesale GPU rate. Shown before every job runs — no surprises after.

most transparent >0

Storage while training

Your dataset and checkpoints sit in cloud storage only for the life of the job.

$0

Idle cost

Nothing runs, nothing is billed, when you have no active job.

Full rate card lands inside your account after sign-up. Figures above are standard rates.

The fine print

Terms, Privacy & Governance

Transparent usage terms, strict non-custodial privacy, and automated billing guardrails designed to protect your compute and funds.

1. Service Scope & Compute Routing

  • Dynamic Orchestration: Vivacious Cloud automatically routes containerized training workloads to cost-optimal GPU instances across authorized cloud broker networks.
  • Zero Inspection: We provide autonomous scheduling, telemetry, and checkpoint staging; we do not author, inspect, or modify your model code or training data.
  • User Ownership & Responsibility: You retain complete intellectual property ownership and legal responsibility for all datasets, model code, and output artifacts you submit.

2. Acceptable Use & Account Integrity

  • Licensing: All submitted scripts, training datasets, and model architectures must belong to you or be properly licensed for your intended workloads.
  • Sandbox Isolation: Scripts must strictly execute within their designated container sandbox and must not attempt network probing, resource scanning, or unauthorized host access.
  • Resumable Training: Checkpoint-capable training scripts are strongly recommended; workloads that support mid-run checkpointing can be seamlessly auto-migrated during node disruptions.

3. Account Suspension & Anti-Fraud Enforcement

  • Zero-Tolerance Abuse: Any attempt to tamper with billing tokens, forge cryptographic permits, bypass metering checkpoints, manipulate wallet balances, or exploit infrastructure triggers immediate account suspension.
  • Model Weight Forfeiture: For accounts terminated due to malicious behavior or billing tampering, access to trained model weights is permanently revoked and locked.
  • Balance Forfeiture: Any remaining deposited wallet balance on suspended accounts is forfeited with zero rights to compute or withdrawal.
Scroll container to read complete terms

Your next training run could already be on a cheaper GPU.

Sign up, add a card, run one command. That's the entire learning curve.

Get started free →