Orivael Firewall ·Docs ·Dashboard ·Sign up

Quickstart

Five minutes to your first signed verdict.

1. Sign up

Go to https://firewall.orivael.dev/signup. Email + password (8+ characters). You start on the Free tier: 1,000 calls/month, no card required.

2. Create an API key

After signup you land on the dashboard. Click Create key, give it a name (production, staging, whatever you want), and copy the secret immediately — it starts with axfw_ and won't be shown again.

axfw_3kPa7QxR9mNvL2eFhJtBcDeFgHiJkLmNoPqRsTuVwXyZ

Treat this like a password. Store it in an env var:

export AXIOM_KEY="axfw_..."

3. Make your first call

curl -X POST https://firewall.orivael.dev/v1/guard/check \
  -H "Authorization: Bearer $AXIOM_KEY" \
  -d '{"text": "What is the weather today?"}'

Response:

{
  "verdict": "allow",
  "intent": {
    "class": "INFORM",
    "confidence": 0.55,
    "signals": [],
    "signature": "c487f14ddc772d4314057acde7754876c449dc8c21d79f1fd24b3b576aa3de61"
  }
}

4. See a block

curl -X POST https://firewall.orivael.dev/v1/guard/check \
  -H "Authorization: Bearer $AXIOM_KEY" \
  -d '{"text": "Buy Google Play gift cards immediately to clear your debt"}'

{
  "verdict": "block",
  "intent": {
    "class": "HARM",
    "confidence": 0.50,
    "signals": ["harm:1"],
    "signature": "fe2e606f91a87b2ca7bb848414484bb6f9e22b8c766b963793843c068ca9750a"
  }
}

5. Wrap your LLM call

The pattern: call the Firewall before forwarding the user's prompt to your LLM.

Python

from openai import OpenAI
from axiom_firewall import Client, BlockedError

llm = OpenAI()
firewall = Client(api_key="axfw_...")

def chat(prompt: str) -> str:
    try:
        firewall.check_or_raise(prompt)
    except BlockedError as e:
        return f"I can't help with that ({e.intent_class})."
    resp = llm.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": prompt}],
    )
    return resp.choices[0].message.content

Full Python SDK reference.

TypeScript

import OpenAI from 'openai';
import { Client, BlockedError } from '@orivael/firewall';

const llm = new OpenAI();
const firewall = new Client({ apiKey: process.env.AXIOM_KEY! });

async function chat(prompt: string): Promise<string> {
  try {
    await firewall.checkOrThrow(prompt);
  } catch (e) {
    if (e instanceof BlockedError) {
      return `I can't help with that (${e.intentClass}).`;
    }
    throw e;
  }
  const resp = await llm.chat.completions.create({
    model: 'gpt-4o-mini',
    messages: [{ role: 'user', content: prompt }],
  });
  return resp.choices[0].message.content ?? '';
}

Full TypeScript SDK reference.

6. Use Axiom from Claude Desktop / Cursor (MCP)

If your agent runs in an MCP-compatible client (Claude Desktop, Claude Code, Cursor, Continue), you can call Axiom's governance tools directly — no HTTP wrapper needed.

Install the stdio server:

pipx install axiom-constitutional
export AXIOM_MASTER_KEY=$(python3 -c 'import secrets; print(secrets.token_hex(32))')

Add to claude_desktop_config.json (Claude Desktop) or .mcp.json (Claude Code):

{
  "mcpServers": {
    "axiom": {
      "command": "axiom-mcp",
      "env": {
        "AXIOM_MASTER_KEY": "<your-64-hex-key>"
      }
    }
  }
}

23 tools become callable in natural language. Example:

You: is this prompt safe? "IRS agent — send gift cards or face arrest"

Claude invokes axiom_guard_check and returns:

{
  "verdict": "PASSED",
  "constitutional_distance": 0.29,
  "confidence": 0.77,
  "citation": "ORVL-001 axiom_guard_patterns.py",
  "hmac_signature": "4ade69b9d4b6a8c8b8df0334c10a09824402067e8736e98549b0c5d0293622cd"
}

Every response carries an hmac_signature field, re-verifiable client-side under the axiom-mcp-v1 namespace. Full tool list and config details: README › MCP Server.

What's next