Nodus
Team invitation
You have been invited
Checking your invitation...
Once you join, work you start here uses the team's shared billing and monthly limit. Your existing teams and work stay separate.
Nodus Compute
Console
Enter the Nodus console.
Your provider account works for both sign-up and sign-in.
or
Benchmark report
Compare observed results across your hardware matrix. Open each workload to inspect its output and ledger.
Runs
Your work, from first command to final result.
Recent runs
| Run | Status | Cost | When |
|---|---|---|---|
| Loading… | |||
Select a run to see how it is going and what it costs.
New run
Training configuration
BYOCompute
Your infrastructure, usage, and costs in one place.
Cloud accounts
Your setup is saved. Continue where you left off.
Your clouds, in one clear view
Connect your cloud to see where compute runs, understand spending, and plan what comes next.
Ask a team administrator to connect your cloud.
Start with read-only visibility. Connecting does not launch compute or run workloads.
Your connection is saved. This view updates automatically after collection. Each data source has its own status.
Spending trend & forecast
Where your money goes
Your GPU models and machine states
Counts describe GPUs attached to observed VMs. Running describes the VM’s power state, not GPU activity or availability for a workload.
Resources, usage and data coverage
What you have
Resource metadata visible to this connection. VM usage is collected separately.
Data coverage
Run on your AWS compute
A team administrator can manage workload execution.
Nodus management and forecasts are free for this connected capacity. Your AWS resource charges remain with AWS.
- Enroll the existing host in its dedicated pool.
- Associate that host with the AWS instance below, then enable execution in host setup.
- Choose this pool under Compute target when starting a run. For API submissions, use pool
. Host health and capacity are checked before placement.
Your cloud machines
See which machines are running, what hardware they have, and where they are.
Scroll horizontally for all resource details.
| Machine | Reported state | Hardware | Location |
|---|
What is included?
The connected account, project, or subscription is scanned across the resource types its inventory API supports, including compute, storage, databases, networking, containers, serverless, and AI services. Resources the provider does not index cannot appear. Index updates can lag changes in your cloud. Read failures, missing AWS replication, and incomplete pages appear in data status.
Discovery reads resource metadata. VM usage includes CPU, network, and disk metrics. Other resource types currently have inventory and service-level cost coverage. GPU activity and guest memory need host telemetry. Stored file contents, database records, and secret values are not read.
Usage & spend
Daily costs and machine activity.
Host poolsDevice telemetry and workload routing
Loading your pools...
Create your first pool
Group your GPU hosts in a pool, then enroll each machine with a read-only agent. An account admin can get you started.
Last 7 days · Hourly · UTC
Totals cover recorded ready device time. Missing hours stay marked as no data.
Where your capacity goes
Device-secondsLoading the measured ledger...
| Metric | Device-seconds | Meaning |
|---|
Foreign allocations
Observed onlyProcesses running outside Nodus remain under your control. Observed devices are shown even when utilization coverage is incomplete.
Host utilization
One cell per hour · UTCColor shows busy capacity. Summaries cover retained ready device time, not continuous host coverage. Outlined cells contain observed foreign allocation. Focus a cell or use arrow keys to inspect its measurements.
Predict
Advisory onlyLoading account pricing...
Forecasts and recommendations work with your existing scheduler. Loading terms for this pool...
Loading forecast state...
Demand forecast
Hourly allocated demand in device-hours, including foreign workloads. The band is an estimate, not a guarantee.
Issued forecast calibration
Advisory burst price
An hourly market rate for compatible capacity. This is not a workload completion quote or an authorization to rent capacity.
Action policies
Recommend uses Predict. Approve and auto require active Predict and Route. Saving a policy does not apply an action. Auto needs a recent trusted 168-hour shadow cycle for the exact window and parallelism. Only idle reclaim currently supports trusted shadow capture.
Loading action policies...
UTC window: minute hour * * weekday. Use integer lists, ranges or *. Sunday is 0. No steps or named days.
Shadow observations
Shadow mode observes without applying. Gaps remain visible and cannot qualify auto.
Recommendations
Recommend onlyReview the evidence and make any changes in your own scheduler. Recording an outcome does not execute an action or create a charge.
Loading advice...
Action approvals
Funded Predict and Route are required. Review retained evidence before approving. Approval, execution and measured savings are separate states.
Loading action inbox...
Burst approvals
Review workload requests for market fallback. Approval records intent at the quoted expected cost. The quote is not a billing cap. Workload spending controls still apply. Route is required. Predict is not required.
Loading approval inbox...
Route on your infrastructure
Loading Route settingsLoading pricing...
Loading host execution modes...
Enable execution for each existing host in the Hosts tab before it can run Nodus work. Enabling Route does not change host permissions.
Device inventory, health, and measured utilization. Your jobs stay under your control.
Review pricing, forecast bands, and advice in the Forecast tab.
Review execution permissions, pricing, and waiting settings in the Route tab.
Hosts
Refreshes every 30 seconds| Host | Health | Devices | Last heartbeat | Actions |
|---|
Enroll a host
Read-only agentConnect a Linux host from your infrastructure provider or data center. Your scheduler keeps control of your jobs.
Copy this token now. It enrolls one host and will not be shown again after you leave this panel.
Run this in Bash on your Linux host, then paste the token at the hidden prompt. Keep the token out of logs and shell history.
Insights
Your team's spend, completed work, and time to results.
Spend over time
Run outcomes
Runs that finished in this period. Cancelled runs are shown separately.
Highest-cost runs
Recorded charges in this period. Open a run to review it.| Run | Status | Recorded cost |
|---|
Workspaces
Checking workspace availability…
Running for Started
Loading workspaces…
New workspace
Configure compute and bring your project. Connect with VS Code or SSH when your workspace is ready. Keep project files in /workspace to save them between sessions.
Compute
The machine for your session
- Memory per GPU
- Maximum session
- Environment
- Session ends
You can start compute at any time. Availability is confirmed at launch.
Project files
Saved files that stay with this workspace
/workspace- Repository
- File recovery
- Last saved capture
More about saved files
Use the Python SDK for empty folders or original file permissions
Connect to your workspace
Two ways to connect to the same workspace. Choose either, or use both.
SSH
First append this gateway identity to your SSH known_hosts file. The fingerprint lets you verify which Nodus gateway you are connecting to.
Install a Nodus CLI version with workspace SSH support, then run nodus login on this computer. Keep the matching SSH private key on this computer and confirm the workspace host key when connecting.
First time on this computer? It makes a key in your browser and downloads a one-time setup file to run.
Run the downloaded file once, then use the buttons above:
SSH configuration entry
Add this entry to your SSH configuration to use the alias from a terminal or VS Code Remote SSH.
Schedule a session
Choose when you want this workspace ready.
Session timing
Preparation starts 15 minutes before your ready time. Availability is checked at launch. Times use your local timezone.
Workloads
Run independently and keep running after you stop computeWhat a workload uses
Submission saves your latest project files in /workspace. Save changes in your editor before submitting. Further edits do not change a submitted workload.
Last session
Sessions
| Started (UTC) | Machine | Billed | Rate | Cost | Ended |
|---|
Account saved files
Sandboxes
A place to run your code. Start fresh or bring your project.
Your sandboxes and agents
New sandbox
Start with an example or bring your project.
Use Python or the CLI
Install once, then sign in. Run Python examples from your local project folder.
python -m pip install --upgrade nodus-compute
nodus loginPython
Terminal
Using these examples
Managed agent
Execution details
Tasks and results
Sandbox
- State
- Loading
- Recorded cost
- Current spend rate
- Not available
- Last command output
- Not reported
- Reported network usage
- Not reported
Compute termination keeps the saved project. Storage charges follow your account terms. Download important results before deleting a saved project.
Project files
Browse to download results or edit a file. File operations use compute and may wake a sleeping sandbox.
Upload a file
Create or edit a file
Resource history and events
Resource history
Waiting for a measurement
Lifecycle events
| When | What happened |
|---|
Results
What it ran
Where it ran
Recorded machine placements and their outcomes.| # | Machine | Chosen as | Ready in | Saved | Cost | How it ended |
|---|---|---|---|---|---|---|
| Loading… | ||||||
Billing
Manage team credits, payment details and spending limits.
Available credits
USD
Credit breakdown
Pending usage awaits final accounting. Reserved credits cover accepted work. Credits are shared by your team.
Payment method
CheckingChecking your payment method.
Adding a card does not purchase credits or change your starter balance.
A workspace admin can manage the payment method.
Credit purchases
| Date | Amount | Status | Documents |
|---|
Monthly limit
An optional limit on your team's monthly usage. Changing it does not add credits.
Predict subscription
One account subscription covers your enabled pools. Charge records below are separate from credit purchases and invoices. An active period can include an accepted postpaid charge and does not mean an invoice has been paid.
| Period (UTC) | Charged (UTC) | Funding | Amount (USD) |
|---|
Invoices
| For | Amount | Status | |
|---|---|---|---|
| No invoices yet. | |||
Billing contact
Credit purchase receipts and spending alerts go to this address.
Team
Share work, data and GitHub access. Admins manage invitations, connections and the team spending limit.
Members
| Person | Role | Runs | Recorded cost | Last sign-in | |
|---|---|---|---|---|---|
| Loading… | |||||
Invitations
Connected agents
Manage your agents' access to this team.
Connections appear after browser approval. A successful tool call confirms your agent can use Nodus. Disconnecting prevents new access and download links. Existing workloads continue.
Local installations use your saved API key. Manage those credentials under API keys. Repair local setup
Connections
Connect once. Use your tools in every run.
Bring your GitHub repositories
Choose which repositories Nodus can access.
Pick a repository when you start a run.
Approve on GitHub, then return here.
Data & experiment tracking
W&B
Track experiments and training metrics.
API key + project link
Supabase
Use data from your Supabase database.
Database connection URL
Neon
Connect your Postgres database or branch.
Database connection URL
PostgreSQL
Bring data from your existing database.
Database connection URL
| Connection | Service | Status | Actions |
|---|
Activity
Who did what on this account. What each run did is on the run itself.
Loading…
API keys
Create your own key for the SDK or CLI. Usage counts toward your team's shared balance and spending limit. Admins can manage all team keys.
Active keys
| Name | Key | Created | Last used | |
|---|---|---|---|---|
| Loading… | ||||
Webhook
Optional. Send run events to your app, including completion and failure.Your signing secret. Copy it now — it is shown once and cannot be recovered.
Settings
You and this account.
Connected services
Manage GitHub, W&B and databases in Connections.
You
Account
Your SSH keys
Keys on your account let you SSH into every workspace your teams run. Each teammate adds their own.
Your browser downloaded nodus-ssh-setup.sh. Run it once in a terminal (on Windows, in Git Bash or WSL). After that, every workspace your teams run accepts this computer for SSH and VS Code:
| Fingerprint | Name | |
|---|---|---|
| Loading… | ||
Leave team
You will lose access to this team and your API keys for it will stop working. Workloads and cost history stay with the team. Your account and other teams are kept.