@standardproofs/cli

Verified answers from your terminal. If you would rather not use a terminal at all, the dashboard at /app does the same thing in the browser — connect your model account once and just type the question. The pipeline runs on StandardProofs infrastructure (Lean + Mathlib, Sage, SymPy); the model calls run on your account.

npx @standardproofs/cli login
export OPENAI_API_KEY=sk-…
npx @standardproofs/cli "A 20kg child is prescribed 300mg of acetaminophen q6h. Is that in range?"

The default setup (theoria56nr) uses OpenAI for every step, so an OpenAI key is all you need. The CLI also supports ANTHROPIC_API_KEY, for setups that route steps to Anthropic. If a key is missing you are told which one, before the run starts rather than part-way through.

  StandardProofs · theoria56nr
  using your OPENAI_API_KEY

  ✓ solver               24s
  ✓ formalizer            8s
  ✓ prover               31s
  gate round 1 · 23 claims · 0 failed

  CERTIFIED   p_de7050c222c9

Your keys

Two different keys, and only one of them is ever stored:

| | what it is | where it lives | |---|---|---| | sp_live_… | your StandardProofs identity | ~/.standardproofs/config.json, mode 600 | | OPENAI_API_KEY (or ANTHROPIC_API_KEY) | your model account | your environment only |

Your model key is read at call time, sent for that one run, and never written to disk or database on either side. Rotate it whenever you like without touching anything here.

Commands

| | | |---|---| | sp login | store your API key (verifies it before saving) | | sp "question" | verify, streaming live | | sp runs | your recent runs | | sp logout | forget the stored key |

--demo runs the real engine on a scripted trajectory with no model calls — useful for checking the integration before involving a real key. --json prints the full result to stdout; -q prints only the verdict.

Exit codes are CI-friendly: 0 certified, 1 declined, 2 error.

Self-hosting

--api https://your-deployment (or STANDARDPROOFS_API) points the CLI at a different deployment.