Simple, honest pricing

Free to start. Pay as you scale. No surprises.

Free

$0/ forever

Get started locally or with the hosted API. No credit card.

Get API key
  • 10,000 authorize() calls / month
  • TypeScript + Python SDK, local or hosted
  • Cryptographic receipts (Ed25519, verifiable offline)
  • Policy console - edit rules without redeploying
  • Policy version history + one-click rollback
  • Session-level aggregate tracking
  • Decision webhooks
  • Usage dashboard
  • LangGraph + OpenAI Agents SDK examples
  • Community support (GitHub)
Most popular

Pro

$79/ mo

For teams that need unlimited calls, audit exports, and priority support.

Get started
  • Unlimited authorize() calls
  • Usage-based pricing
  • Evidence Vault - CSV/JSON audit exports for SOC2 / EU AI Act
  • Priority email support
  • SLA: 99.9% uptime

Enterprise

Custom

For companies that need dedicated support and the deepest compliance story.

Contact us
  • Everything in Pro
  • KMS/HSM-backed receipt signing (Roadmap)
  • Extended Evidence Vault retention + custom exports
  • SAML / SCIM (Roadmap)
  • Dedicated SLA + support contract

Performance

Sub-10ms or developers leave

Any authorization check that adds visible latency to agent execution gets removed. Mizara evaluates synchronously, stateless, with no database query mid-request. Real numbers from the production engine:

Mizara stateless engine~1.2ms
Traditional DB authorization check~85ms
System prompt guardrail400ms–1.2s*

* system prompt is non-deterministic and cannot be relied on for enforcement

Cryptographic receipts

Every decision is provable

Every authorize() call - including ALLOWs - produces a SHA-256 hash of the full decision payload, signed with Ed25519. Verification only needs the public key from GET /api/v1/public-key - not a live call back to Mizara's server, and not trust in Mizara staying online or honest.

This is what a SOC2 / EU AI Act auditor actually tests for. An append-only log file is not the same thing.

{
  "status": "DENY",
  "evaluation_metadata": {
    "triggered_rule_id": "rule_block_prod_delete",
    "policy_version": 3,
    "execution_time_ms": 1.182
  },
  "enforcement": {
    "action_halted": true,
    "user_facing_error": "Production deletion requires approval"
  },
  "cryptographic_receipt": {
    "id": "rcpt_fe0fc435eec5ec2f",
    "hash": "6a29337ce524cc2d...",
    "signature": "kQ3mP8vXz1yT...",
    "algorithm": "ed25519",
    "public_key": "9fLxK3vQ2mNp..."
  }
}

Common questions

What counts as an authorize() call?

One call to the POST /api/v1/authorize endpoint. The free tier includes 10,000 per month. Local SDK usage with your own policy file is always free and unlimited.

Can I use the local SDK without an account?

Yes. npm install @mizara/sdk and pip install mizara both work with a local policy.json file, no API key, no account. The account is only needed for the hosted API.

When will Pro pricing be available?

Shortly. We're validating the schema with early design partners first. Sign up for early access and you'll be first to know.

What's the difference between a log file and a cryptographic receipt?

A log file can be edited. A cryptographic receipt is a SHA-256 hash of the full decision payload signed with Ed25519, an asymmetric algorithm - so anyone can verify it offline with just the public key, without trusting Mizara's server. For SOC2 and EU AI Act compliance, this distinction matters.

Can I verify a receipt without calling Mizara?

Yes. Fetch the public key once from GET /api/v1/public-key (no auth required), then use verifyReceipt() from either SDK to check any receipt offline. This is what makes the receipt an audit artifact rather than a claim you have to trust.

Can I prove what a policy looked like when a specific decision was made?

Yes. Every policy edit is versioned and kept immutably. Each receipt records the exact policy_version active at decision time, and the dashboard's Evidence Vault export joins the two so a past decision and the rule that produced it are never separated.