Self-serve API that scores sign-ups, logins and transactions for bots, account takeover and abuse, from a free plan with 1,000 Risk API calls.
Castle is the pick for a SaaS or consumer app team that wants login and sign-up abuse blocked without a sales cycle. Pro includes 40,000 risk checks. Shops that need payment fraud scoring, and companies that need mobile SDKs or certifications, should look elsewhere.
Self-serve account takeover and bot scoring with published per-request prices and a free plan.
Policies, lists and up to 18 months of event history live inside Castle, and its published prices make a rival easy to price against.
Main trade-off
Mobile coverage and procurement-grade assurances sit behind a $4,000-a-month Enterprise plan.
| Plan | Monthly price | What it includes |
|---|---|---|
| Free | Free | $5 of API usage a month: 1,000 Risk requests or 5,000 IP lookups, Web SDK, bot, abuse and ATO scores, 1 request per second, 3 days of data exploration, 3 seats, 1 environment, documentation support |
| Pro | $200/mo | $200 of API usage, then $0.005 per Risk request and $0.001 per IP lookup, 5 requests per second, 7 days of data exploration, 5 seats, 2 environments, Chat and email support |
| Enterprise | $4,000/mo | From $4,000 a month, MTU or volume pricing, iOS and Android SDKs, emulator, jailbreak and root detection, Full IP database downloads, SLA and dedicated integration support |
Free $0 a month with $5 of API usage (1,000 Risk or Filter requests, or 5,000 IP lookups) for evaluation, no overages. Pro $200 a month with $200 of API usage (40,000 Risk requests or 200,000 IP lookups), then pay as you go at $0.005 per Risk or Filter request and $0.001 per IP lookup, no contractual minimum. Enterprise from $4,000 a month with MTU or volume pricing, mobile SDKs and an SLA. Unused API budget does not roll over.
Castle is a developer-first fraud API that returns bot, abuse and account takeover scores for each sign-up, login or transaction, using device, IP, behaviour and email signals. It stands out for self-serve plans with published per-request prices, starting with a free tier of 1,000 risk checks a month.
Bot, abuse and takeover scores
Each sign-up, login or transaction event returns three separate scores that policies can act on.
IP intelligence
Flags proxies, Tor, VPNs and residential proxies, sold also as a stand-alone lookup API at $0.001 per IP.
Device fingerprinting
Recognises devices across visits and spots headless browsers and tampering, with multiple-accounts-per-device signals.
Policies and lists
Allow, challenge or deny in real time, with trusted and blocked lists of users, devices or any custom attribute.
AI agent identification
Names supported AI agents and browser automation tools in a dedicated agent object, with operator and category.
Phone call detection
Detects a user on a phone call during login or a transaction, a common sign of social engineering.
Most users expect these and they are not currently supported.
| Tool | Price from | Free plan | Runs on | Rating |
|---|---|---|---|---|
| $200/mo | Yes | Web | 3.6 | |
| Quote | No | Web | 3.9 | |
| $699/mo | Trial | Web | 3.8 | |
| -- | Trial | Web · iOS · Android | 4.1 |
Proprietary
Pro is $200 a month and includes $200 of usage; after that each Risk or Filter request costs $0.005 and each IP lookup $0.001, with no minimum contract. Enterprise starts at $4,000 a month and can be priced per monthly tracked user instead. Unused usage does not carry over to the next month.
Yes, the Free plan costs $0 and includes $5 of usage a month, which covers 1,000 risk checks or 5,000 IP lookups. It is meant for evaluation: it has no overages, one request per second and three days of data, so production traffic needs Pro.
An IP lookup returns connection intelligence for one address through a simple call. A Risk or Filter request needs the full Castle integration and returns device, behaviour and risk-score data about the user, which is why it costs five times more.
Only on Enterprise. The iOS and Android SDKs, and emulator, jailbreak and root detection, are not included in Free or Pro, which cover web apps.
It can identify supported AI agents and browser automation tools and returns each one's name, operator and category in a separate agent object. Your policy can then treat a known shopping assistant differently from an anonymous script.
Castle is an identity and fraud risk platform that stops abuse at sign-up, login and during transactions. Developers add its SDK to a web app, send events to its Risk API, and security or fraud analysts then watch and act on attacks in Castle's dashboard. You can sign up and start on a free plan without talking to sales.
The mechanism is three scores on every event: a bot score, an abuse score and an account takeover score, built from device fingerprinting, IP intelligence, behaviour and email signals. Its IP data flags proxies, Tor, VPNs and residential proxies, the home-internet addresses attackers rent to look like ordinary users, and it marks disposable email domains. Recent changelog entries added detection of a user who is on a phone call during a sensitive action, a sign of a scammer coaching a victim, and identification of named AI agents and browser automation tools. Policies then allow, challenge or deny each request in real time, with lists, custom signals and webhooks.
It fits engineering and security teams at software companies and consumer apps that want account takeover and fake sign-up protection they can switch on themselves, with a bill they can predict from published per-request prices.
Castle does not score payments or offer a chargeback guarantee, so online stores still need Stripe Radar or Signifyd. Mobile SDKs and mobile tamper checks are Enterprise only. Large platforms under heavy bot attack can compare the managed, enterprise-only Arkose Labs or Sift, and teams that want email and phone footprint checks look at SEON. See the AI security tools list.
Product facts on this page (pricing, plans, integrations, compliance) are read from Castle's own pages and documentation. The verdict is Hokai's reading of them, not the vendor's claim.