Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.sault.ai/llms.txt

Use this file to discover all available pages before exploring further.

For agents with shell access. One package, one pairing flow, one config file.

​
Option 1: ask the agent to pair itself

Paste this into your agent: 
Set up SAULT for paid HTTP requests.

1. Run: npx -y @sault.ai/cli@latest login
2. Show me the approveUrl and verificationCode it prints to stderr.
3. Wait — the command long-polls. I will approve in the browser.
4. Once it returns success, read the manual: npx -y @sault.ai/cli@latest skill
   and save it to this project's rules file.
The CLI mints an agent key, writes it to ~/.sault/config.json (mode 0600), and stays paired across sessions.

​
Option 2: run it yourself

npx -y @sault.ai/cli@latest login
When the CLI prints an approveUrl and a short verificationCode, open the URL, sign in at app.sault.ai, and type the verification code from your terminal into the approve page. Approve, and the key is saved. Verify: 
npx -y @sault.ai/cli@latest status

​
Option 3: paste an existing key

If you already have an agent key from the console: 
npx -y @sault.ai/cli@latest pair --key sault_ak_...
Skips the browser flow entirely.

​
How the agent learns

The CLI ships a SKILL.md — a manual written for the model, not the human. It covers the three main scenarios (single paid call, inspecting a service, running a published skill), the JSON envelope every command returns, and the retry / idempotency rules. Print it: 
npx -y @sault.ai/cli@latest skill
Save it to your project’s agent rules file (CLAUDE.md, .cursor/rules, etc.) so the manual is in context next session.

​
Output shape

Every command emits a JSON envelope on stdout. Prompts (URLs, verification codes, status updates) go to stderr so JSON pipelines stay clean. 
{ "ok": true,  "data":  { ... } }
{ "ok": false, "error": { "code": "...", "message": "..." } }