Skip to main content

CLI reference

The paresium command is the one-shot lane: one command, one governed result, a machine-readable receipt. It is built for scripts, automation, and AI agents that use Paresium Clara as a tool. Add --json to any command for structured output.

paresium --help
Usage: paresium [OPTIONS] COMMAND [ARGS]...

Options
--json Emit machine-readable JSON output.
--profile <str> Configuration profile name. [default: dev-gcp]
--version Print version and exit.

Commands
console Open the interactive REPL console.
init First-run setup: write config, login, verify connectivity.
check Run a governed check against Paresium Clara.
account Show identity and credit balance.
usage Show recent settlement / credit usage.
knowledge Add and manage your compliance rules and playbooks.
records Look up saved records.
llm Direct model access (metered).
bench Open the visual console (TUI).
… auth · traces · policy · playbook · persona · audit · report

First run

paresium init # write config, sign in, verify connectivity
paresium account # confirm your identity and credit balance

Run a governed check

The core offering: a safe, reviewed answer with a record behind it.

paresium check --message "What is our refund policy for enterprise plans?"

Add --json to get the full receipt (decision, reasoning, record id) for a script or agent to consume. Pick the model with --model clara-dev (only the beta model is live today).

Add knowledge (the agent lane)

Teach Clara a compliance rule or playbook. Every publish quotes a price first and is checked so it can never weaken a required safety rule.

paresium knowledge list # what Clara already follows
paresium knowledge import ./refund-policy.md --structure compliance
paresium knowledge review # see the pending draft
paresium knowledge review publish # commit it (metered)
Commands
list List your knowledge docs.
versions Version history for one doc (every publish is versioned).
import Import document(s) (metered, quoted first) or JSON items.
delete Retire a doc (Clara stops following it; versions are kept).
scope Change who a doc applies to (creates a new version).
review Approve and publish imported documents.

Prove a new rule landed by running a compliance check that the rule governs; the result names the rule and points back to the document. See Teach Clara.

Direct model access

Send your own instructions straight to the model. This lane is always metered (never test billing), so you spend real credits.

paresium llm quote --message "Draft a friendly refund email" # price first
paresium llm exec --message "Draft a friendly refund email" # run it

Records and usage

paresium records get <record_id> # the full saved record for one run
paresium usage # recent credit settlement

Every record carries which model produced the answer (clara-dev / clara-0.1-beta1 today), so the audit trail is complete.