Use the console

The console at console.nodus-compute.ai shows the same objects the CLI and the Python SDK work with. Every page has the command that does the same thing, so you can move between them at any time.

Sign in

1. Open the console and sign in with your email and password, or with GitHub or Google.
2. New accounts confirm their email with a 6-digit code.
3. If someone invited you, the console shows the invite first. Accept it to join their organization.
4. Otherwise create your organization, then select a project in the sidebar.

Your first organization receives starter credit once your email is verified.

Sign in the CLI from the browser

Run nodus login. The CLI opens the console, which asks which organizations the CLI may use and names the key it creates, for example cli-laptop-2026-09-30. Choose Authorize and return to your terminal. On a machine without a browser, run nodus login --device and enter the code it prints on the console’s device page.

If you have no organization yet, authorizing creates one for you. If someone invited you, both pages show the invite first: accept it, and the CLI signs in to their organization.

Find your work

• URLs name the page: /runs, /jobs/lora-fine-tune/logs, or /billing/overview. The organization and project come from your selection in the sidebar. Refresh and Back preserve the selection for that page.
• A short link opened directly uses the recipient’s selected organization and project. Select the matching organization and project when sharing resource links. Older links containing an organization and project still open that exact scope and then show the shorter address. Different browser tabs can keep different selections.
• Switch organization or project from the top bar.
• Choose Search in the sidebar to find objects, pages and actions. Type job/ or sb/ and a name to jump straight to an object.

Overview and BYOC

Overview shows active work, failures, endpoint readiness and recent completed work in the selected project. Training and other parent runs appear once rather than repeating their underlying jobs. Open an item to see its existing logs, status and results. Creation links stay available above the activity.

Choose hardware when creating a job, training run or workspace. Training runs your training command; Workspaces provide an interactive development environment. Both use the resources you select in their creation forms.

BYOC (Beta) manages your own cloud accounts, pools and machines. Its routing switches control whether eligible jobs and training runs use your cloud accounts and whether they can overflow to Nodus capacity. Open Cloud accounts for connections or Pools & machines for enrolled capacity.

Under Inference → On-demand, choose Automatic (Indra) for routed requests or select a specific model from the available catalog. The request panel preserves that model selection in the request and code example. Endpoints remains the view for the project’s persistent endpoints. RL environments remain a main navigation item.

Follow runs live

Lists and detail pages update as things change, without reloading. The note Live · 12s ago shows the time since the last update. Choose it and Pause live updates to freeze what you are reading; Resume applies the changes at once. Running work shows its cost so far, marked with ≈ between updates.

Read agent and inference results

Agent answers and on-demand inference responses render Markdown automatically, including headings, lists, tables, links and code blocks. Formatting updates as inference text streams in.

Copy the command

Choose Copy CLI command from the page actions menu to copy the command with the project and organization spelled out:

nodus describe job/lora-fine-tune -p default --org research

View YAML in the actions menu shows the live object as a manifest you can download and apply elsewhere.

Create something

Creation follows Configure work → Choose resources → Review & launch for jobs, workspaces, sandboxes, endpoints, agents, agent runs, and runnable RL examples. The final action creates the resource; moving between steps does not launch it. Going back preserves your configuration. Manifest-based forms keep YAML, CLI, and Python in Manifest views, using the same configuration as the form.

Jobs, workspaces, and RL examples use the same Automatic placement / Choose hardware chooser. RL examples start with the training template’s resource preset; choose Set requirements or exact hardware to override it. Their review uses a server dry run before launch. These flows save a browser-local draft and reuse the request key after an uncertain submission. If a price changes, review the refreshed estimate.

Endpoints choose a catalog model and policy limits; agents choose their model access and team limits; sandboxes choose CPU resources. These APIs do not accept GPU placement constraints, so their resource step shows only supported settings. Inference’s playground keeps its lightweight prompt-and-send flow for repeated requests.

When credits run out

If your balance cannot fund the next renewal, running work stops at its funded edge. The project Overview then shows Paused for funds, grouped by what adding credits does:

• Resumes automatically: jobs and agent runs continue by themselves once a hold fits.
• Needs a start: workspaces stay stopped until you choose Start (or Start all).
• Wakes on next use: sandboxes start again on the next command, file request or preview.
• Rejecting requests: inference endpoints answer with 402 until funded.
• Failed imports: choose Retry import on the volume.
• Needs a higher limit: work that reached its own spend limit resumes only when you raise that limit.

Choose Add credits on the panel to top up.

Appearance

Choose System, Light or Dark from the account menu. The theme follows you across the console and the docs. The console always uses Comfortable spacing. Account contains Profile and SSH keys.

Product analytics

When configured, the signed-in console collects page-view analytics. There is no analytics switch in Account. Browser Do Not Track and Global Privacy Control signals are respected. Automatic input capture and session replay are disabled. Prompts, logs, secrets and file contents are not collected as analytics events.

Choosing hardware

Start a job, training run or workspace, then choose its resources in the creation form. Automatic placement lets you set GPU count, minimum memory and region; Nodus selects matching capacity. Alternatively, use Choose hardware to browse GPU or CPU configurations, search by name, and filter by GPU count, region and availability. Select a card, then review your configuration before launching. That choice pins the offering and region in the manifest, including whether it is interruptible. Unavailable hardware queues until matching capacity is available, subject to the request’s queue timeout. Use Choose hardware to change that selection. YAML and SDK views carry the same placement constraints.

Moving between offerings

Overview summarizes active work, failures, ready endpoints, and recent results. Its Create work menu opens the same job, workspace, RL, inference, and agent flows as the sidebar. Expand the failure list or outcome history when you need more detail. Work sections stay together; models, datasets, storage and BYOC sit under Resources.

Runs includes jobs, training, pipelines, sweeps, and agent runs. Filter by workload type or choose Failed to investigate a problem. Reinforcement learning opens its training and evaluation history, with separate views for environments and training templates. A training run opens the same detail page from either section. Open an environment and choose Configure training or Configure evaluation beside a runnable example. The example retains its environment version, model, parameters and task split; SDK and CLI access remain available.

Models & datasets lists references found in training runs, including model revisions and dataset volume revisions and paths. Each reference links back to the runs that use it; volume-backed references also open their storage. Missing versions are labeled Not pinned. This view is a reference index, not a checkpoint registry.

In the inference playground, Use in agent carries a tool-capable model into a new agent definition. Create endpoint carries the selected model into endpoint setup. Neither action sends a request or launches work until you submit the next form. On Workspaces, use the row’s Connect or Open action to reach its connection controls.

Compare hardware and investigate failures

Select Compare on up to three hardware cards to compare GPU count, VRAM, CPU, RAM, region, interruptibility, and current availability. L40 and L40S remain distinct models. Their peak dense BF16 figures link to NVIDIA’s specifications and are not predictions of application throughput. Your exact hardware choice remains in the manifest when you move between creation steps.

Overview’s Needs attention rows include the reported failure cause and recovery guidance, with links to logs and configuration. Jobs and workspaces also offer Review & run again, which opens a new draft for review. It does not retry or allocate compute immediately.

Training outputs link to the source run, input model revision, dataset volume and revision, compiled execution, and any declared export volume. Models and datasets cross-link through their actual training references. A ready endpoint with a tool-capable catalog model offers Use in an agent, preserving the endpoint reference. Custom checkpoint deployment is not currently supported by the inference endpoint API, so no Deploy action is shown for checkpoint files. The console does not infer a checkpoint revision from a volume’s mutable latest revision.