CLI

Use rundial to authenticate, select a project and work with runs from a terminal.

Set RUNDIAL_API_KEY to your project ingestion key before running these commands.

pip install rundial
rundial target use --workspace default-workspace --project default-project
rundial run list

If a command prints No stored credentials found, set RUNDIAL_API_KEY in that shell. The list shows training runs in the selected project. Use rundial auth whoami to check access and rundial auth logout to clear stored credentials.

Profiles are stored in ~/.config/rundial/config.toml. Secrets use the system keyring when available. --allow-plaintext-store explicitly permits the fallback at ~/.config/rundial/secrets.json. Protect this file if you enable it. The default endpoint is http://127.0.0.1:8787; RUNDIAL_ENDPOINT, RUNDIAL_WORKSPACE and RUNDIAL_PROJECT can override profile settings.

Global options

Pass these before the command. Every command also accepts -h or --help.

Option	Purpose
--profile <name>	Select a profile.
--endpoint <url>	Override the API endpoint.
--format human|json	Select output format.
--non-interactive	Disable prompts.
--verbose, --quiet, --no-color	Control output.
--allow-plaintext-store	Permit local plaintext secret storage if keyring is unavailable.

Commands

Options listed with --scope also accept --scopes; repeat them to provide multiple values. <run-id>, <name> and eval run IDs are positional arguments. Use command --help for usage.

Command	Options and action
auth login	--api-key, --workspace, --project, --allow-plaintext-store: save credentials.
auth whoami, auth logout	Show the principal or clear credentials.
auth key create	--workspace, --project, --scope, --agent, --label, --expires-at-ms: create a key.
auth key list	--workspace, --project: list keys.
auth key revoke	--key-id (required), --workspace, --project: revoke a key.
auth token create or token create	--label, --scope, --expires-at-ms, --expires-in-days, --use: create a personal token and optionally use it.
auth agent create	--workspace, --name (required): create an agent.
auth agent list	--workspace, --status active|revoked: list agents.
auth agent revoke	--workspace, --agent (required): revoke an agent.
auth agent key create	--workspace, --project, --agent (required), --scope, --label, --expires-at-ms: create an agent key.
auth agent key list	--workspace, --agent (required): list agent keys.
auth audit list	--workspace, --agent, --actor-kind user|api_key|agent|system, --limit: list audit events.
target ls	--workspace: list targets.
target use	--workspace, --project: set the default target.
workspace ls	List accessible workspaces.
project ls	--workspace: list projects.
run start	--workspace, --project, --name, --kind training|agent|eval: start a run.
run list	--workspace, --project, --status, --search, --limit, --order asc|desc, --order-by runId|startedAtMs|updatedAtMs|runName: list runs. --status can be repeated.
run status <run-id>	--workspace, --project: inspect a run.
run finish <run-id>	--state completed|failed|aborted, --reason, --finished-at-ms: finish a run.
metrics tail <run-id>	--workspace, --project, --keys, --points, --follow, --interval-seconds: print metric points.
metrics export <run-id>	--workspace, --project, --keys, --format json|csv, --out (required), --start-ms, --end-ms, --page-size: export metrics.
logs tail <run-id>	--workspace, --project, --min-level trace|debug|info|warn|error|fatal, --text, --lines, --follow, --interval-seconds: print logs.
logs export <run-id>	--workspace, --project, --format json|csv, --out (required), --start-ms, --end-ms, --page-size: export logs.
evals diff <baseline-id> <candidate-id>	--workspace, --project, --scorer, --limit, --fail-on-regression: compare eval runs.
profile ls, profile use <name>, profile delete <name>	List, select or delete profiles.
profile create <name>	--endpoint, --activate, --no-activate: create a profile.
config show	--all: show resolved configuration or all profiles.
init	--api-key, --workspace, --project, --name, --skip-run-start, --allow-plaintext-store: set up a target and optionally start a run.
doctor	--offline: inspect local configuration without network checks.
doctor-capabilities	Show server CLI capabilities.

Next steps

• Members and API keys
• HTTP API