@sault.ai/cli

Terminal client for AI agents that pay HTTP services. Wraps the SAULT agent-payments API into a small set of commands an LLM can drive end-to-end: pair the device, call any URL (auto-handles 402), check vault balances and spending, browse the marketplace, run published skills.

Use sault login to pair the current environment with a SAULT agent. When funds are needed, sault topup returns a vault-scoped dashboard link for the user.

Install

npm i -g @sault.ai/cli

Or run ad-hoc without installing:

npx @sault.ai/cli@latest <command>

Requires Node.js 18 or newer.

Quickstart

sault login                    # opens a browser, mints + saves an agent key
sault status                   # confirms the key is configured and reachable
sault topup                    # opens the dashboard funding flow if balance is low
sault request <url>            # makes the request — pays automatically on 402

Configuration

sault login writes the API key to ~/.sault/config.json with mode 0600. Override at the process level with environment variables:

| Variable | Purpose | |--------------------------|--------------------------------------------------------------| | SAULT_API_KEY | Use this key instead of ~/.sault/config.json | | SAULT_BASE_URL | Point the CLI at a non-default API host (e.g. local backend) | | SAULT_IDEMPOTENCY_KEY | Reuse a fixed Idempotency-Key for sault request |

sault config prints the current configuration with the key redacted.

Commands

Identity / setup
  status      Show config + ping the API
  login       Browser device-flow pairing
  pair        Save an existing API key directly (--key sault_ak_...)
  topup       Print a vault-scoped dashboard top-up URL
  doctor      Check this CLI's version against the published latest/min
  config      Show current redacted config

Marketplace
  discover           Browse paid services
  skill show         Print a published skill (raw markdown)
  skill prepare      Readiness check before running a skill
  skill              Print this CLI's bundled SKILL.md

Payments
  request <url>      HTTP request — auto-handles 402 challenges
  check <url>        Probe a URL for 402 requirements (no payment)
  poll <url>         Poll an async paid job

Inspect agent state
  whoami             Identity, policy, caps
  balance            Vault balances
  spending           Daily spend vs. cap
  history            Recent payments

Run sault --help for the full reference.

Output

All commands except sault skill show and sault skill emit a JSON envelope on stdout:

{ "ok": true,  "data":  { ... } }
{ "ok": false, "error": { "code": "...", "message": "..." } }

Human-readable prompts (URLs to open, verification codes, status updates) go to stderr so JSON automation on stdout stays clean.

For AI agents

The package ships with a SKILL.md manual — agent-readable instructions covering the three main scenarios (single paid call, inspecting a service, running a published skill), response shapes for each command, and the retry / idempotency rules. Print it with:

sault skill

License

MIT — see LICENSE.