CLI
Import local data, repeat training configurations, compare evaluations and retrieve outputs from your terminal or CI.
The tensorant command connects your terminal and scripts to your Tensorant organization. Import local files, freeze a dataset version, launch and watch training, compare evaluations, run experiments and download the results. Long-running work runs on Tensorant and continues when your terminal closes.
| What you want to do | Guide |
|---|---|
| Install, sign in and import your first file | CLI quickstart |
| Find a command or download an output | Command reference |
| Save configurations and enforce quality in CI | Automation and configuration |
| Choose an organization, manage profiles or revoke access | Authentication and profiles |
Access and spending
The CLI uses the same personal grants as MCP, with your current organization role and plan limits checked on every request. Sign in in your browser, or supply a personal token for CI. Model API keys from the application API do not authenticate CLI commands.
| Access | CLI workflows |
|---|---|
| Read | Inspect projects, sources, versions, jobs, runs, evaluations and experiments; download outputs |
| Build | Also import and review examples, create versions, evaluate existing endpoints and cancel background jobs |
| Full | Also create and run experiments, start training or trials, retry runs, and manage test deployments; owners can publish models |
Paid starts return a spending preview before they launch. Interactive use asks for confirmation; automation must explicitly pass --yes. Use --preview to inspect the request without launching. Endpoint evaluations can incur provider charges at Build access. Training limits and experiment estimate ceilings are described in Automation and configuration.
The console remains the place for connections, memberships, billing, API keys and organization settings. Use the MCP tools when you want an agent to choose and carry out workflow steps conversationally.
Need a hand? Visit troubleshooting