The AI workspace that keeps working
Steer agents in Chat. Watch progress on the Feed. Compound knowledge in Wiki. Heartbeat continues between sessions — with a Daytona sandbox when the work needs a computer.
What the workspace did while you were away.
Wrote map/home and posted an artifact to the Feed.
Next tick: research note, then ask if blocked.
Publish the research note to wiki/research, or keep iterating?
Models and compute behind the factory
Google
Anthropic
xAI
Meta
AWSEverything in the workspace, nothing extra to babysit
Six primitives. Humans push the levers. Agents do the useful work — and leave a trail you can audit.
Feed
The workspace home. Heartbeats, artifacts, and asks land here so autonomy is something you can actually see.
Chat
Steer the orchestrator. Set goals, answer questions, and point the next tick — you stay in control.
Wiki
NeuralHub Memory. Durable notes, maps, and research that compound across sessions instead of vanishing in a thread.
Heartbeat
Wake hourly, daily, or on demand. One meaningful move — or a clear ask if the workspace is stuck.
Sandbox
A real computer for the workspace. Daytona is the primary environment — install tools, run scripts, keep artifacts.
Workflows
Curated skill packs you enable per workspace. Recipes mount into the sandbox and show up in chat and heartbeat.
How the factory loop works
Autonomy only counts when you can see it. NeuralHub makes progress visible, durable, and resumable.
- 01
Steer
Open Chat, set a goal, or drop a note on the Feed. You decide the direction.
- 02
Work
Heartbeat or a chat tool loop takes one useful step in the sandbox or wiki.
- 03
Remember
Progress is written to the Feed and Wiki so the next session starts with context.
- 04
Continue
Resolve asks, run heartbeat again. The workspace compounds instead of resetting.
A living factory — not another chat box
Chat is how you steer. The product is the loop around it: memory that lasts, a schedule that wakes the workspace, and a computer when work needs one.
- Visible autonomy
If it did not land on the Feed, it did not happen.
- Human gates
Stuck work becomes an ask — you resolve it, then run again.
- A real computer
Install whatever the job needs inside an isolated Daytona sandbox.
Usage-based credits. No monthly seats.
Pay for inference and sandbox time. Credits never expire. Sign in with Google — no card required to create an account.
OpenRouter models inside the workspace. Billed in credits as you chat and run heartbeat.
Daytona environments billed by the second. A default box is about $0.17/hour.
- Home workspace on first sign-in
- Feed, Chat, Wiki, Queue, Workflows
- Scheduled or manual heartbeat
- Primary Daytona sandbox
- Credits never expire
- No monthly platform fee
Frequently asked questions
- Is NeuralHub a model API?
- No. It is a workspace product. You sign in, land in a home workspace, and run the factory loop — Feed, Chat, Wiki, Heartbeat, Sandbox. Models are the engine, not the product.
- What happens when I close the tab?
- Heartbeat can wake the workspace on a schedule (hourly, daily, or off). It takes one meaningful step or posts a clear ask on the Feed for you to resolve.
- Do I need to provision infrastructure?
- No. Your home workspace is created on sign-in. The primary sandbox (Daytona) starts when work needs a computer — scripts, installs, or skill packs.
- How does billing work?
- Prepaid credits. You pay for model inference and sandbox time. There is no monthly seat fee, and credits do not expire. Sign in with Google to start — no credit card required to create an account.
Open a workspace and run the loop
Sign in with Google. Your home workspace is ready — Feed, Chat, Wiki, and Heartbeat.