DEVELOPERS · THE ENGINE AS ONE HTTP CALL

Give your AI a decision engine.

Chat models give opinions. This endpoint runs the full solvemax pipeline — clarify, widen the field, locked-weight scoring, 5,000 seeded simulated futures, red team, blind second-pass audit — and returns the complete verdict slip as JSON plus a shareable URL. The pipeline itself never leaves our servers; you integrate a black box that ships receipts. 1 verdict credit per call, from your account balance (failed runs auto-refund). Keys live in your dashboard.

60-SECOND START

curl

curl -X POST https://solvemax.solveright.ai/api/v1/verdict \
  -H "Authorization: Bearer smx_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "decision": "Two job offers. A: BigTech, $310k, 4 days in office. B: Series B startup, $205k + 0.4% equity, remote. Mortgage, first kid due in November.",
    "auto": true
  }'

With auto: true the engine asks itself the clarify questions and takes its own recommended answers (each is confidence-tagged in the response). To answer them yourself, pass "answers": [{"q": "...", "a": "..."}]. Optional "mode": DECIDE · COMPARE · GRILL. Responses stream nothing — they return the finished slip in 60–180s, so set your client timeout ≥ 300s. Full schema: openapi.yaml.

CLAUDE · CLAUDE CODE · CURSOR · ANY MCP CLIENT

MCP server (npx, zero install)

The bundled MCP server exposes one tool, get_verdict. Add to Claude Desktop / Claude Code / Cursor config:

{
  "mcpServers": {
    "solvemax": {
      "command": "npx",
      "args": ["-y", "solvemax-mcp"],
      "env": {
        "SOLVEMAX_API_KEY": "smx_live_...",
        "SOLVEMAX_BASE_URL": "https://solvemax.solveright.ai"
      }
    }
  }
}

Then just say: "use solvemax to grill this launch plan." The model hands the decision to the engine and gets the slip back — podium, dual verdict, kill criteria, verification — instead of improvising an opinion. (Source for the thin client is in the repo under integrations/mcp; publish it to npm as solvemax-mcp once, or point command at the local file.)

CLAUDE CODE

/solvemax slash command

# save as .claude/commands/solvemax.md in any repo
---
description: Run the solvemax decision engine on a decision
---
Run this decision through the solvemax engine and summarize the slip:
!curl -s -X POST https://solvemax.solveright.ai/api/v1/verdict -H "Authorization: Bearer $SOLVEMAX_API_KEY" -H "Content-Type: application/json" -d '{"decision": "$ARGUMENTS", "auto": true}'
CHATGPT

Custom GPT Action (5 minutes)

  1. ChatGPT → Explore GPTs → Create → Configure → Actions → Import from URL: https://solvemax.solveright.ai/openapi.yaml
  2. Auth: API Key · Bearer → paste your smx_live_… key.
  3. Instructions: "For any hard decision the user describes, call get_verdict and present the slip: headline, podium, dual verdict, exit plan, verification. Include the slip URL."

Gemini (Gems with API tools) and Grok follow the same pattern with the same OpenAPI file — or fall back to the curl recipe in any tool that can run shell.

SHELL

One-liner for your dotfiles

solvemax() { curl -s -X POST https://solvemax.solveright.ai/api/v1/verdict \
  -H "Authorization: Bearer $SOLVEMAX_API_KEY" -H "Content-Type: application/json" \
  -d "{\"decision\": \"$*\", \"auto\": true}" | jq -r '.slip.headline, .url'; }
# usage: solvemax "quit my job in March to go full-time on my SaaS? $40k runway, 9 customers"
RULES OF THE ROAD

Limits & guarantees

  • 1 credit per verdict · failed runs are auto-refunded · credits_remaining rides on every response.
  • 30 verdicts/hour per key · 5 active keys per account · decision text ≤ 20,000 chars.
  • Your decisions stay private; API-created slips are visible only to your account unless you publish them.
  • Same engine, same determinism: identical inputs reproduce identical numbers.