REST API · Streaming

NSFW Chat & Roleplay API — uncensored LLM with persona memory

Q: How do I authenticate?

Bearer token in the Authorization header. Generate keys from your dashboard — a sandbox key for testing and a production key with rate limits and credit metering. Keys are revocable, can be scoped per-domain, and support per-key IP allowlists for server-to-server use.

Q: Is the API really uncensored?

Yes for adult, romantic, explicit and fantasy content between consenting adults. We block CSAM by default (and report it), and require age-gate + consent headers for explicit content. Outside those hard limits, the models respond as tuned — no refusal layer on adult roleplay.

Q: Do you support streaming?

Yes — SSE streaming is the default, with 50ms TTFB p50 and token-level delivery. Non-streaming JSON responses are also supported. Both follow the OpenAI Chat Completions wire format so existing streaming clients work unchanged.

Q: What's the rate limit?

Shared-fleet default is 60 concurrent requests per key and 5M tokens per 24h. Volume tiers lift both. Dedicated endpoints have no hard rate cap — limited only by the GPU(s) in your cluster and your configured queue depth.

Q: Which models are supported?

Llama 3.1 70B and 8B (NSFW-tuned), Mistral Large 2, Nous Hermes 70B, Pygmalion 12B, PsyMed RP 13B, plus any custom LoRA you upload. Model selection is per-request via the model field — you can A/B test within a single conversation if needed.

Q: Can I self-host?

We offer bring-your-own-GPU deployments where we operate vLLM/Triton on your infrastructure under a managed-services contract. Pure code licensing is also available for larger operators who already run their own inference fleet. Most customers start on the shared API and move to dedicated when volume justifies it.

Q: Is the API 2257-compliant?

We provide the building blocks: per-request record hooks, optional content moderation exports, age-gate header enforcement, and CSAM pre-filtering with reporting. We're not a 2257 records keeper for your business — you remain the custodian — but our logs and exports are designed to make audit conversations survivable.

Q: How is content moderated?

CSAM is blocked pre-inference and post-inference by classifier + perceptual hash, and reported per legal requirement. Other moderation is configurable: you can disable filters entirely for adult roleplay, enable a light safety net, or run your own classifier on the stream. We expose moderation flags via webhooks and per-conversation exports.

Q: Can I get a dedicated endpoint?

Yes. Dedicated H100 clusters start at $4,500/mo for a single-tenant deployment with isolated context cache, custom SLA, dedicated model routing and 24/7 on-call. Multi-region failover is available for enterprise tiers. Talk to sales for a quote tailored to your throughput and latency targets.

A production chat & roleplay endpoint that streams uncensored responses token-by-token from NSFW-tuned Llama 3.1, Mistral Large, Pygmalion and custom fine-tunes. Drop-in replacement for OpenAI Chat Completions — without the content refusals. Persona memory, multi-turn RP, system prompts, LoRA hot-swap and 50ms time-to-first-byte on streaming.

Get API Access View documentation

50ms

Streaming TTFB p50

150+

NSFW-tuned models

99.9%

Uptime SLA

<$0.40

Per 1K output tokens

TL;DR

Endpoint: POST https://api.nsfwcoders.com/v1/chat/completions — OpenAI-compatible request/response schema, SSE streaming by default, Bearer token auth, optional LoRA and persona_id parameters.

Models: Llama 3.1 70B, Llama 3.1 8B, Mistral Large, Nous Hermes, Pygmalion 12B, PsyMed RP, NSFW-tuned Llama — plus custom fine-tunes you can hot-swap per request.

Pricing: pay-as-you-go at $0.40 / 1K output tokens (cheaper on volume), credit-metered with no monthly minimums. Dedicated H100 endpoints available for $4,500/mo.

SLA: 99.9% uptime with multi-region failover; 50ms streaming TTFB p50 and 800ms non-stream p50 on the shared fleet.

Overview

A chat API that doesn't refuse half your prompts

Most chat APIs — OpenAI, Anthropic, even open-source providers behind a safety filter — silently refuse the moment a conversation turns adult. Your roleplay breaks, your character goes off-script, and your users churn. Our NSFW Chat & Roleplay API is built specifically for the adult case: uncensored LLMs fine-tuned on roleplay corpora, served with no refusal layer, with persona memory that survives long sessions.

It's OpenAI-compatible at the wire level — same chat/completions endpoint, same streaming SSE format, same tool-calling shape — so most existing SDKs work with a base-URL change. You bring your characters, system prompts and LoRA weights; we bring the GPUs, the inference engine, and the operational maturity of running adult chat for our own companion platforms.

Who uses it: companion-app teams replacing OpenAI mid-flight after a refusal storm, cam sites adding AI-driven persona chat to creator pages, dating apps prototyping AI matches, roleplay-marketplace founders, and studios generating adult scripts at scale. If you've ever shipped a chat feature and watched it break on adult content, this is the endpoint you needed from day one.

What makes it different from generic APIs: uncensored by default (not 'uncensored' as a marketing label), NSFW-tuned Llama and Pygmalion models you won't find on OpenRouter, persona memory primitives built into the API, and a provider that actually wants adult traffic on its infrastructure.

OpenAI-compatible /v1/chat/completions endpoint — drop-in SDK swap

Streaming SSE with 50ms TTFB p50 and token-level delivery

150+ NSFW-tuned models incl. Llama 3.1, Mistral Large, Pygmalion, Nous Hermes

Persona memory, system prompts, tool/function calling, JSON mode

Hot-swap LoRA per request — your fine-tunes or ours

Optional moderation hooks, CSAM pre-filter, 2257 record exports

NSFW Chat & Roleplay API — uncensored LLM — engineered by NSFW Coders

60-day delivery

prototype to production

Product showcase

What your users actually see

A live-grade interface built on the same components we ship to production — designed for retention, monetization, and scale.

NSFW chat & roleplay API streaming response dashboard by NSFW Coders — Token-by-token streaming for multi-turn adult roleplay at 50ms time-to-first-byte.

Who it's for

Who calls this API

AI companion apps

Replace OpenAI when refusals started costing retention — same SDK, no migration tax.

Cam & creator platforms

Add AI persona chat alongside live creators to fill dead-air hours and monetise off-peak.

Dating apps

Prototype AI matches, icebreakers and conversation prompts that don't trip safety filters.

Roleplay marketplaces

Multi-character RP with persona swaps, scene state and long context windows.

Adult content studios

Generate dialogue, scripts and scenarios at scale for video and audio pipelines.

Affiliate funnel builders

Power NSFW chatbots that convert paid traffic without the API pulling the plug mid-campaign.

Why NSFW Coders

Why this API vs OpenAI or Replicate

/ 01

Actually uncensored

No refusal layer on adult, romantic, explicit or fantasy content — by design, not by omission.

/ 02

NSFW-tuned models

Pygmalion, Nous Hermes, NSFW Llama fine-tunes — not available on OpenRouter or Anthropic.

/ 03

Adult-payment-friendly

We accept adult use cases and adult-business payments; no surprise deplatforming mid-campaign.

/ 04

Built-in compliance

Optional CSAM pre-filter, age-gate headers, 2257 record hooks — for processor and audit conversations.

/ 05

Operated GPU fleet

Our own H100/A100 clusters with vLLM paged attention and Triton — we run the same fleet for our apps.

/ 06

3 years of NSFW chat

Production-hardened on 4B+ tokens/month of real adult traffic — not a fresh wrapper.

Authority & track record

Three years running uncensored LLMs

in production adult traffic — not a wrapper around a wrapper

NSFW-tuned model fleet

Llama 3.1 70B, Mistral Large, Pygmalion and Nous Hermes fine-tuned on adult roleplay corpora — served hot on H100/A100 with vLLM paged attention.

Battle-tested on real traffic

The same chat endpoint powers our own companion and white-label platforms — 4.2B+ tokens/month of NSFW dialogue with 99.9% uptime.

Adult-business-friendly

We accept adult use cases by default. No silent ban risk, no surprise deprecation — your API key works for as long as you pay.

Compliance helpers built in

Optional CSAM pre-filter, age-gate headers, 2257 record hooks and per-conversation moderation exports make audit conversations survivable.

99.9% SLA + dedicated endpoints

Multi-region failover with optional single-tenant H100 clusters for volume customers who need predictable latency and isolated context caches.

What's included

API features what ships in the endpoint

OpenAI-compatible schema

Same /v1/chat/completions request/response shape, same SSE streaming, same tool-calling.

Streaming SSE

Token-by-token delivery at 50ms TTFB p50 — no buffering, no first-token delays.

Persona memory

Per-conversation memory_id keeps character state, history and preference graph across calls.

LoRA hot-swap

Pass lora_id per request to load character-specific fine-tunes in <500ms cold-start.

Python + Node SDKs

Typed SDKs with streaming helpers, retry/backoff and persona state management.

Signed webhooks

Conversation-end, moderation-flag and credit-low events delivered with HMAC signatures.

Credit metering

Per-request token accounting with pre-paid balance, soft and hard limits, and refund on inference errors.

Rate limits & quotas

Per-key concurrent request caps and 24h token quotas with rolling usage dashboards.

Idempotency & retries

Idempotency-Key header deduplicates retries safely; backend-level retry on transient 5xx.

Process

Integration flow — step by step

Get API key

Test in sandbox

Hit the sandbox endpoint with cURL or our Postman collection; verify streaming, persona_id and LoRA loading.

Migrate from OpenAI

Change base URL from api.openai.com to api.nsfwcoders.com — most SDKs work as-is.

Go live

Swap to production key, set rate limits, configure webhooks and credit alerts; monitor dashboards.

Tune personas & LoRAs

Upload character fine-tunes or use ours; A/B model selection and temperature per cohort.

Scale to dedicated

At volume, move to a single-tenant H100 cluster with isolated context cache and custom SLA.

GPU-optimised pipeline

Architecture

How the platform is wired

NSFW Coders engineers the full stack — from model serving and GPU orchestration to the application layer, payments, moderation and analytics. Every layer is built to be audited, scaled, and swapped without re-platforming.

Model layer

Stable Diffusion, Flux, Pony, custom LoRA & 3D — served via vLLM / ComfyUI / Triton.

API gateway

Key-auth, rate limits, credit metering, signed webhooks — multi-tenant.

Application layer

Next.js / React, realtime chat, in-app feed, creator studio.

Data & safety

Postgres + Redis + vector store, CSAM scanning, age gates, 2257 logs.

Discuss your architecture

4.2B+

Tokens served monthly

150+

NSFW-tuned models

50ms

Streaming TTFB p50

99.9%

Uptime SLA

Tech & stack

Models & serving stack under the endpoint

Models

Llama 3.1 70B (NSFW-tuned)Llama 3.1 8B (NSFW-tuned)Mistral Large 2Nous Hermes 70BPygmalion 12B v2PsyMed RP 13BCustom fine-tunes via LoRA

Serving

vLLM with paged attentionTriton Inference ServerRedis persona cacheKubernetes GPU autoscaleH100 + A100 fleetMulti-region failover

Stack

OpenAI-compatible RESTSSE streamingPython SDKNode.js SDKOpenAPI 3.1 specPostman collectioncURL quickstart

Use cases

What teams build on it

AI companion chat

Girlfriend/boyfriend personas with memory, mood and explicit-option tiers.

Example: Replaced OpenAI after 30% refusal rate; tokens dropped 60% in cost.

Multi-character RP

Group roleplay with scene state, NPC personas and turn-taking.

Example: Roleplay marketplace running 8-character scenes at 4K context.

Cam-site persona chat

Creator AI twin chats fans off-peak.

Example: +22% creator-page revenue from off-hours engagement.

Adult script generation

Long-form scene scripts and dialogue at scale.

Example: Studio generating 1,200 scripts/month for video pipeline.

Dating-app icebreakers

In-app AI matches that don't trip safety filters.

Example: Pilot launched with 18+ age gate; no API warnings.

NSFW chatbot funnels

Affiliate chatbots that convert paid traffic.

Example: Chatbot on popup traffic; ROAS 1.6 at $0.05 CPC.

Comparison

NSFW Coders chat API vs alternatives

Feature	NSFW Coders	Generic OpenAI/Replicate*	Open-source self-hosted
Adult content allowed	By default	Refused	Depends on your filter
NSFW-tuned models	Pygmalion, NSFW Llama	Not available	DIY fine-tunes
Streaming TTFB	50ms p50	200–400ms	Depends on infra
Operational burden	Zero	Zero	Full-time SRE
Adult-payment friendly	Yes	Often refused	N/A
Compliance helpers	2257 + CSAM + age-gate	None	DIY
Cost at scale	$0.40 / 1K out	Higher + filters	GPU capex + ops
SLA	99.9% multi-region	Provider SLA	Self-managed
Time to first call	Minutes	Minutes	Weeks of setup

* Generic providers may change content policies without notice; OpenAI/Anthropic refuse adult content by default.

Monetization

Built to make money on day one

Subscription tiers, token packs, pay-per-message, pay-per-view media, creator splits, affiliate payouts and tipping — wired into adult-friendly payment processors so revenue is never blocked by a sudden account freeze.

Subscription + credits

Hybrid billing that lifts ARPU without choking free-funnel conversion.

Creator economy

Multi-creator payouts, revenue share, content locks and PPV media.

Affiliate & referrals

Tracking links, first-touch attribution, automated payouts.

High-risk payments

Segpay, CCBill, Paxum, Verotel — with fallback routing.

Per-token credit metering and quota dashboard for the NSFW chat API

Pricing

API pricing, pay-as-you-go or dedicated

Pay-as-you-go

$0.40 / 1K output tokens

Shared H100/A100 fleet. No monthly minimums. Credit-metered with prepaid balance; cheaper on volume tiers.

150+ NSFW-tuned models
Streaming SSE + tool calling
Persona memory & LoRA hot-swap
99.9% SLA on shared fleet
Python + Node SDKs

Get API access

Questions, answered

Quick answers to the questions founders ask us most about this service.