# Features

> **Sentinel v2 (beta):** This page documents Sentinel v2, a beta rewrite that may change before the stable release. It does not apply to Sentinel v1, the current stable version: see the [Sentinel v1 documentation](/docs/sentinel.md).

Sentinel combines several independent detection mechanisms into one decision. This page shows where each one acts, what turns it on, and which endpoint exposes it, so you can pick the pages that matter for your integration.

## Where each feature acts

A widget submission passes three stages. Each feature acts in one or more of them.

1. **Before the challenge.** The API key's [security group](/docs/sentinel/v2/concepts/security-groups/) evaluates its rules against the request. Rules can deny it based on [threat intelligence](/docs/sentinel/v2/features/threat-intelligence/) (`threat`, `threatVerdict`), the [agent](/docs/sentinel/v2/features/bots-agents/) that signed it (`agent`), the client's [penalty](/docs/sentinel/v2/features/rate-limiting/) or [risk](/docs/sentinel/v2/concepts/scoring/#risk), its country, network, and headers. Matching `set` rules choose the settings for the next two stages, and [rate limits and header entropy](/docs/sentinel/v2/features/rate-limiting/) are counted.
2. **The challenge.** `GET /v2/challenge` issues a [proof-of-work](/docs/sentinel/v2/features/proof-of-work/) challenge with the algorithm and cost the rules chose. Rules can add a [code challenge](/docs/sentinel/v2/features/code-challenge/) or turn on [Human Interaction Signature](/docs/sentinel/v2/features/human-interaction-signature/), which scores how the user interacted with the page before the challenge is issued.
3. **The submission.** `POST /v2/verify` checks the solution, then the [classifier](/docs/sentinel/v2/features/classifier/) scores the request and the form fields the widget sends: [spam, profanity, and injection](/docs/sentinel/v2/features/text-analysis/) in the text, [phishing URLs](/docs/sentinel/v2/features/phishing/), [similarity](/docs/sentinel/v2/features/similarity/) to your training data, and, when a group opts in, the [AI decider](/docs/sentinel/v2/features/ai-decider/). The verdict is signed into `verificationData` for your backend. See [Verifying submissions](/docs/sentinel/v2/integrate/verification/).

Your backend can also call most detection features directly, without a widget, with a `restricted` or `full` API key. See [Backend detection APIs](/docs/sentinel/v2/integrate/backend-apis/).

## Feature map

| Feature | Acts in | Turned on by | Direct API |
| --- | --- | --- | --- |
| [Proof-of-work](/docs/sentinel/v2/features/proof-of-work/) | Challenge | Always on; `pow.*` set fields tune it | `GET /v2/challenge` |
| [Code challenge](/docs/sentinel/v2/features/code-challenge/) | Challenge | `codeChallenge.enabled` set field | — |
| [Human Interaction Signature](/docs/sentinel/v2/features/human-interaction-signature/) | Challenge | `his.mode` set field | `POST /v2/challenge` |
| [Rate limiting & penalty](/docs/sentinel/v2/features/rate-limiting/) | Every request | `[rate_limit]` (global, on by default); `rateLimit.*`, `headerEntropy.*`, `penalty.add` set fields | — |
| [Classifier](/docs/sentinel/v2/features/classifier/) | Submission | Always on in `POST /v2/verify`; `classifier.*` set fields tune it | `POST /v2/classifier` |
| [Spam, profanity & injection](/docs/sentinel/v2/features/text-analysis/) | Submission | Classifier text and injection categories, when text is sent | `POST /v2/classifier` |
| [Similarity & training data](/docs/sentinel/v2/features/similarity/) | Submission | Campaign detection on by default; training data in the account; `[similarity] provider` for semantic matching | `POST /v2/similarity` |
| [AI decider](/docs/sentinel/v2/features/ai-decider/) | Submission | `[classifier.ai]` provider plus `classifier.ai` set field | — (runs inside the classifier) |
| [Threat intelligence](/docs/sentinel/v2/features/threat-intelligence/) | Every request, submission, admin logins | Threat sources and entries; `threat` rule conditions; `[auth] block_threats` (on by default) for logins | `POST /v2/threat` |
| [Phishing detection](/docs/sentinel/v2/features/phishing/) | Submission | `[classifier.phishing]` list | `POST /v2/phishing` |
| [Email analysis (EML)](/docs/sentinel/v2/features/email-analysis/) | Backend only | `[eml]` | `POST /v2/eml` |
| [Bots & AI agents](/docs/sentinel/v2/features/bots-agents/) | Before the challenge | `[web_bot_auth]`; `agent` rule condition | — |
| [Lookups & inspect](/docs/sentinel/v2/features/lookups/) | Debugging, enrichment | `[ip_resolver]`; `[inspect]` | `POST /v2/ip`, `/v2/language`, `/v2/user-agent`, `/v2/time-zone`; `GET /v2/inspect` |

Credential types differ by endpoint: widget endpoints take a `public` key, detection endpoints a `restricted` or `full` key. See [API keys & access levels](/docs/sentinel/v2/concepts/api-keys/).

## What a new security group does

A security group without rules still protects a widget:

- Every challenge is proof-of-work with `PBKDF2/SHA-256` at cost `5000`.
- `POST /v2/verify` classifies the request (and the form fields, when the widget sends them) and signs the verdict.
- The global per-IP rate limit, `[rate_limit] global` (default `100/1m`), applies.
- The code challenge, Human Interaction Signature, rule rate limits, header entropy, and the AI decider are off.

To turn the rest on without writing rules by hand, apply the **Recommended** preset in the admin UI. It blocks IPs with a threat verdict of `block`, rate-limits each device, limits header entropy, and gives clients with high risk or penalty a code challenge and a harder proof-of-work. See [Security presets](/docs/sentinel/v2/policies/presets/#recommended).

## Where to start

- Protecting a form: [Proof-of-work](/docs/sentinel/v2/features/proof-of-work/), [Classifier](/docs/sentinel/v2/features/classifier/), and [Spam, profanity & injection](/docs/sentinel/v2/features/text-analysis/), then [Writing security rules](/docs/sentinel/v2/policies/security-rules/) to escalate on risk.
- Moderating user content from your backend: [Classifier](/docs/sentinel/v2/features/classifier/), [Similarity & training data](/docs/sentinel/v2/features/similarity/), and [AI decider](/docs/sentinel/v2/features/ai-decider/).
- Screening inbound email: [Email analysis (EML)](/docs/sentinel/v2/features/email-analysis/).
- Debugging a decision: [Lookups & inspect](/docs/sentinel/v2/features/lookups/) and [Scores, risk & penalty](/docs/sentinel/v2/concepts/scoring/).

## In this section

- [Proof-of-work](/docs/sentinel/v2/features/proof-of-work/): Sentinel's proof-of-work algorithms, including memory-hard Argon2id and scrypt, and how security-group rules set their cost.
- [Code challenge](/docs/sentinel/v2/features/code-challenge/): An image or audio code that security-group rules can require for suspicious requests, and its accessibility features.
- [Human Interaction Signature](/docs/sentinel/v2/features/human-interaction-signature/): How Human Interaction Signature (HIS) scores interaction signals, its modes, replay detection, and its limitations.
- [Rate limiting & penalty](/docs/sentinel/v2/features/rate-limiting/): Global and per-rule rate limits, Ephemeral Device Keys, header entropy checks, and the penalty score.
- [Classifier](/docs/sentinel/v2/features/classifier/): How the classifier scores a submission across email, device, IP, location, text, similarity, and other categories into a verdict.
- [Spam, profanity & injection detection](/docs/sentinel/v2/features/text-analysis/): Sentinel's deterministic multilingual spam filter, profanity detection, and injection rules for submitted text.
- [Similarity & training data](/docs/sentinel/v2/features/similarity/): Match submissions against your labeled training data, detect near-duplicates and campaigns, and manage training suggestions.
- [AI decider](/docs/sentinel/v2/features/ai-decider/): Let an AI model review submissions during classification: providers, per-group settings, data sent, and failure behavior.
- [Threat intelligence](/docs/sentinel/v2/features/threat-intelligence/): Block or allow IPs from threat sources and from manual or automatic entries, and use threat data in rules and the classifier.
- [Phishing detection](/docs/sentinel/v2/features/phishing/): Detect known phishing URLs in submitted text and check URLs with the phishing API.
- [Email analysis (EML)](/docs/sentinel/v2/features/email-analysis/): Analyze raw inbound email messages: authentication results, attachments, malicious hashes, and EML rules.
- [Bots & AI agents](/docs/sentinel/v2/features/bots-agents/): Identify bots, crawlers, and AI agents with Web Bot Auth signatures, the agent rule condition, and presets, and control their access.
- [Lookups & inspect](/docs/sentinel/v2/features/lookups/): Look up IP, language, user-agent, and time zone information, and inspect how Sentinel evaluates a request.
