# API Reference

> The complete Nosmai Moderation API: methods, result shapes and enums, shared across every platform.

> For AI agents: the complete documentation index is at https://nosmai.com/llms/moderations.txt

Product: Nosmai Moderation
Group: reference
Source: https://nosmai.com/docs/moderations/api-reference/

This reference lists the methods, result shapes and enums the SDK exposes. Method names vary slightly per platform (for example iOS `initialize(licenseKey:)` vs Android `init(context, key)`), so the exact signatures live in the [platform guides](/docs/moderations/ios/). The result fields and enum values below are the same on every platform.

## Methods

| Method | What it does | Returns |
| --- | --- | --- |
| `initialize(licenseKey, [models])` | Validate the license and load the requested models. Call once at startup, off the main thread. | Init result: `{ success, error }` |
| `initializeText()` | Load the text-moderation model. Call after `initialize`. | `bool` |
| `analyzeImage(image)` | Moderate a single image (objects + NSFW). | Visual result |
| `analyzeVideo(path, frameIntervalMs)` | Sample a recorded clip every `frameIntervalMs` and aggregate. Not available on web. | Video result |
| `moderateText(message)` | Moderate a chat message or comment. | Text result |
| Live camera | Stream frames and receive a result per frame. Native / React Native push frames or render a preview view; web calls `startLive` with a `<video>` element. | Visual result per frame |
| `setThreshold(category, value)` | Override an object-class confidence bar at runtime. Not adjustable on web. | void |
| `setNsfwThreshold(class, value)` | Override the NSFW block (`explicit`) or warn (`sexy`) bar. Not adjustable on web. | void |
| `setPerformanceMode(mode)` | Live only: how often the heavier object detector runs. NSFW always runs every frame. | void |
| `shutdown()` | Release the models and free memory. | void |

## Visual result (image and live frame)

| Field | Type | Description |
| --- | --- | --- |
| `isUnsafe` | bool | True if any object is flagged, or NSFW is `block` |
| `detections` | list | Flagged objects, each `{ category, confidence }` |
| `nsfw` | enum | `safe`, `warn` or `block` |
| `nsfwScores` | object | `safe`, `sexy`, `explicit` (0 to 1) |
| `rawScores` | object | Best object score per class, for tuning |

## Text result

| Field | Type | Description |
| --- | --- | --- |
| `blocked` | bool | True if the message should be blocked |
| `layer` | enum | Which layer decided: `none`, `blocklist` or `classifier` |
| `category` | enum | `safe`, `profanity`, `toxic`, `hate`, `harassment` or `threat` |
| `score` | number | `1.0` for a blocklist hit, the model probability otherwise |
| `matchedWord` | string | The banned word that matched (blocklist hits only) |

## Video result

| Field | Type | Description |
| --- | --- | --- |
| `isUnsafe` | bool | True if any object was flagged, or any frame was NSFW `block` |
| `categories` | list | Distinct object categories flagged across the clip |
| `flags` | list | Per flagged frame: `{ timestampMs, category, confidence }` |
| `framesAnalyzed` | number | How many frames were sampled |
| `nsfw` | enum | Worst NSFW verdict across the clip |

## Enums

| Enum | Values |
| --- | --- |
| Object category | `weapon`, `drug`, `cigarette`, `alcohol` |
| NSFW verdict | `safe`, `warn`, `block` |
| NSFW class (for `setNsfwThreshold`) | `explicit` (block bar), `sexy` (warn bar) |
| Text category | `safe`, `profanity`, `toxic`, `hate`, `harassment`, `threat` |
| Text layer | `none`, `blocklist`, `classifier` |

> [!NOTE]
On native platforms enums are typed values (for example `NosmaiCategory.weapon`); on React Native and web they cross as plain strings you can compare directly.