Own your AI creator stack

YouTube analytics, AI generation, and GPU control in one dashboard.

PastaWater lets you analyze your channel, generate content on your own GPUs or cheap rented ones, and run a coding agent against the same hardware. Dashboard analytics come free. The AI stack is what you pay for.

Works with your own hardware, rented Vast.ai nodes, and the `pw-agent` CLI.

Analytics

Free on every tier

Compute

BYOG or rent cheap GPUs

Agent

Same fleet powers pw-agent

Control room

One fleet, four surfaces

Live

Surface 1

Analytics

Surface 2

Generate

Surface 3

Playground

Surface 4

pw-agent

Channel intelligence

Views

1.8M

Subs

+12.4k

Revenue

$9.1k

GPU fleet

Home RTX 4090

94% util

12.2 GB / 24 GB

Vast A5000

$0.43/hr

Provisioned

Friend 3090

Idle

Ready to borrow

pw-agent

fleet terminal

$ pipx install pw-agent

$ pw-agent -p "refactor auth flow"

routing request to qwen3-coder on Home RTX 4090…

done: patch applied, tests started, stream-json enabled

Product stack

Built for creators who want ownership, not another per-token bill.

PastaWater is not a single AI endpoint behind a markup. It is the operational layer for creator analytics, generation workflows, rented or self-hosted GPUs, and a local coding agent.

Included free

YouTube analytics dashboard

Multi-channel sync, revenue and view tracking, quota visibility, and publishing workflows in one place.

8-step flow

AI creation pipeline

Go from topic to script, TTS, captions, music, prompts, images, and final assembly without leaving the dashboard.

No lock-in

Bring your own GPU

Connect Linux, Mac, or WSL nodes in minutes and only pull the models you actually enable.

1-click rental

Vast.ai built in

Browse, rent, and auto-onboard cloud GPUs from the same control plane when you need temporary horsepower.

Automated output

AI manager campaigns

Schedule repeatable “create 3 shorts a week about X” campaigns with cost previews before execution.

Model selection

Playground for every model

Try image, video, audio, music, and LLM models in the browser before you commit them to a pipeline.

pipx install pw-agent

pw-agent CLI

Run a Claude-Code-style coding agent on your own hardware with tools, memory, RAG, and MCP output.

Private fleet sharing

Friends and GPU leasing

Share spare capacity privately with your circle, lease GPUs by the hour, and keep it off the public marketplace.

Manual to silent

Operator modes

Switch between Operator, Copilot, and Autopilot depending on how much control you want at each stage.

How it works

Connect channel. Pick GPU. Generate.

The landing page story needs to be simple: connect your data source, choose compute, then let PastaWater operate the pipeline with as much control or automation as you want.

Modes

Operator

Manual picks for every model and step

Copilot

Suggested route with human approval

Autopilot

Silent execution when speed matters

01

Connect your channel

Pull analytics, comments, and publishing context into PastaWater so the system understands your actual audience.

02

Pick your compute

Bring a home GPU, attach a friend’s spare box, or spin up a Vast.ai rental without juggling extra tooling.

03

Generate, review, and ship

Run the full pipeline, resume any step, and send the finished asset to storage or scheduled YouTube publishing.

Product surfaces

Show the product, not just generic feature claims.

These are the screenshot targets worth designing around: `/generate`, `/gpu-setup`, `/playground`, and the `pw-agent` terminal.

/generate

Eight-step cockpit

Topic, script, voice, captions, music, prompts, images, final assembly.

Focus 1

Resume any stage

Focus 2

Swap models mid-run

Focus 3

Autopilot for unattended jobs

/gpu-setup

Fleet control

Heartbeat, VRAM, utilization, temperature, and hourly cost across local and rented nodes.

Focus 1

No port-forwarding

Focus 2

Selective model pulls

Focus 3

Live node telemetry

/playground

Model playground

Try loaded image, video, audio, music, and text models before routing them into a production workflow.

Focus 1

Fast A/B testing

Focus 2

Browser-native checks

Focus 3

Better model selection

pw-agent

Terminal agent

Code against the same GPU fleet with slash commands, MCP compatibility, and stream-json output.

Focus 1

Local coding agent

Focus 2

Memory + RAG search

Focus 3

PyPI installable

Compute strategy

Run on the hardware you have. Rent the hardware you do not.

That distinction is the landing page moat. PastaWater keeps creator workflows and developer workflows attached to the same fleet instead of sending everyone back to closed inference vendors.

Home or office GPU

Zero token tax

Use the hardware you already own and keep data close to your workflow.

  • Linux, Mac, or WSL onboarding
  • Runs over normal Wi-Fi
  • Only downloads enabled models

Vast.ai rental

Cost-per-hour, up front

Spin up cheap cloud GPUs with provisioning and node onboarding handled automatically.

  • GPU choice inside dashboard
  • Volumes and hourly cost shown
  • Becomes part of your fleet on boot

Hybrid stack

Mix and match

Keep some steps local, burst others to rented hardware, and route the same stack into pw-agent.

  • Same control plane
  • Shared storage outputs
  • Fits creator and developer workflows

Friends and leasing

Private GPU sharing without turning your hardware into a public marketplace listing.

Borrow

Spin up spare capacity from someone you trust.

Lease

Track credit balance and per-hour usage privately.

Route

Use the same shared boxes inside the generation pipeline or pw-agent.

Developer hook

Run your own coding agent against the same fleet.

pw-agent is the part of the product that makes PastaWater feel wider than creator software. It turns the GPU fleet into a terminal-native coding assistant with MCP compatibility and stream-json output.

  • Claude-Code-style local workflow with file edits, tests, and commits.
  • Fleet routing for coder models like qwen3-coder and gemma-family variants.
  • Slash commands, memory, codebase RAG, voice mode, and LoRA prep.

Install

$ pipx install pw-agent

$ pw-agent --setup

Capabilities

serve-mcp
stream-json output
fleet-backed tool routing
codebase memory

Live session

$ pw-agent serve-mcp

mcp tools exposed: generate_image, generate_video, chat_with_fleet…

$ pw-agent -p "add pricing telemetry"

opened 6 files, applied patch, queued tests

stream-json: { status: "running", route: "home-4090" }

Pricing

Analytics stay included. Your stack grows when your fleet grows.

All tiers include YouTube analytics, BYOG support, Vast integration, pw-agent, and access to the full model stack. Jobs on your own hardware avoid per-token AI costs. Vast rentals are billed at cost.

Free

Try before buying

$0

/ month

  • Analytics only
  • 1 channel
  • Best way to explore the dashboard
Compare plan

Starter

Hobby creators

Soon

  • BYOG enabled
  • 1 GPU
  • All content tools with community support
Compare plan

Creator

Weekly uploaders

$9

/ month

  • 3 channels
  • 3 GPUs
  • AI Manager and priority Vast onboarding
Compare plan

Pro

Studios and serious channels

Popular

$29

/ month

  • 10 channels
  • 10 GPUs
  • Shared fleets and stronger support
Compare plan

Studio

Agencies

$79

/ month

  • Unlimited channels and GPUs
  • White-label options
  • Analytics API and onboarding help
Compare plan

FAQ

The two core questions are always compute and control.

That is where the positioning has to stay sharp: you can operate on your own GPUs, and you can still keep manual control when you need it.

Do I need my own GPU?

No. PastaWater works with rented Vast.ai boxes too, and those rented nodes show up in the same fleet view as local hardware.

What do I actually pay for?

Analytics are included. You pay for platform tiers and any rented GPU time, while jobs on your own hardware avoid per-token API costs.

Is pw-agent a separate product?

It is part of the stack. The same fleet that powers generation can also run the CLI coding assistant and expose MCP-compatible tools.

Can I keep control over the workflow?

Yes. Operator mode is fully manual, Copilot suggests, and Autopilot runs end to end when you want the system to act without prompts.

Final call

Own the creator stack instead of renting every step of it.

Start with free analytics, connect a GPU when you are ready, and let PastaWater handle the operational mess between channel intelligence, generation, and developer tooling.