Micro
$3 / month
- 100 requests / day
- 250 additional free-model requests / day
- 3,000 Bits / day
- Models up to 4x requests
An OpenAI-compatible AI API for roleplay that remembers the story.
Reach Claude, GPT, Gemini, DeepSeek, Kimi and a free open-weight tier through one endpoint and one key with long-context memory, community lorebooks, reusable prompts and workflow plugins on top.
Free account · no credit card · works with JanitorAI, SillyTavern, ChubAI, Agnai and RisuAI
from openai import OpenAI
client = OpenAI(
base_url="https://composite.lucidity.sh/v1",
api_key="cmp-...",
)
resp = client.chat.completions.create(
model="anthropic/claude-opus-5",
messages=[{"role": "user", "content": "Continue the scene."}],
)
# Lots of models/span>
Also speaks
Anthropic /v1/messages
Gemini
OpenAI Responses
Mari
- Models in the catalog
- 400+Models, one key
- Free-tier models
- 9+Free every day
- API formats accepted
- 5API formats spoken
- Cost to start
- $0To create an account
Popular AI models for roleplay and apps
One API key reaches the whole catalog, so moving a story from a cheap everyday model to a flagship one is a change to a single field. These are the models people reach for most; each page lists its context window, live pricing and how it behaves in long roleplay.
| Model | Model ID | Context | Input / M |
|---|---|---|---|
| Claude Opus 5 | anthropic/claude-opus-5 |
1M | ... |
| Claude Sonnet 5 | anthropic/claude-sonnet-5 |
1M | ... |
| GPT-5.5 | openai/gpt-5.5 |
1M | ... |
| Gemini 3.7 Flash | google/gemini-3.7-flash |
1M | ... |
| DeepSeek V4 Pro | deepseek/deepseek-v4-pro |
1M | ... |
| Kimi K3 | moonshotai/kimi-k3 |
1M | ... |
| GLM 5.2 | z-ai/glm-5.2 |
1M | ... |
| MiniMax M3 | minimax/minimax-m3 |
1M | ... |
| Qwen3.8 Max | qwen/qwen3.8-max |
1M | ... |
| Mistral Large | mistralai/mistral-large-2512 |
262K | ... |
Prices are read live from the API when this page loads. They are the catalog rate an ordinary request pays; a prompt prefix the provider serves from cache is billed lower. Browse the full model list to compare every model by price and context length.
Free AI models, no credit card
Every Discord-verified account gets a daily allowance on a tier of open-weight models. They cost nothing, they take the same OpenAI-compatible request as the paid models, and the same key switches between them.
- Llama 3.3 70B
- DeepSeek V4 Flash
- GLM 5.2
- Kimi K2.6
- GPT-OSS 120B
- Gemma 4 31B
- Nemotron 3 Ultra
- Mixtral 8x22B
- Step 3.7 Flash
Free models need nothing but an account. There is no Discord requirement and no invite to wait for. Every account gets a daily allowance; credits or a subscription raise it.
Keep long roleplays affordable
Long roleplay is an input-token problem: every turn resends the whole conversation, so the input price is what actually decides the monthly bill.
The pricing chart compares Composite input pricing against other providers.
Prefer a flat monthly bill? The plans below cover a fixed daily allowance across the whole catalog instead of per-token billing.
Monthly plans
Plans sell requests, not tokens: a fixed daily count that refills every day, so the bill does not move with the length of your chat history. Credits work the other way — you top up once and spend it per request. You can use both.
Basic
$5 / month
- 200 requests / day
- 500 additional free-model requests / day
- 5,000 Bits / day
- Models up to 8x requests
Plus
$10 / month
- 500 requests / day
- Unlimited free-model requests
- 10,000 Bits / day
- Models up to 50x requests
Pro
$15 / month
- 1,000 requests / day
- Unlimited free-model requests
- 20,000 Bits / day
- Any model in the catalog
Ultra
$20 / month
- 2,000 requests / day
- Unlimited free-model requests
- 30,000 Bits / day
- Any model in the catalog
Enterprise
$50 / month
- 10,000 requests / day
- Unlimited free-model requests
- 50,000 Bits / day (the balance cap)
- Any model in the catalog
A request is one call to a 1x model. Expensive models cost more than one, and the models page shows each one's multiplier. A plan only draws models up to its own limit from its daily requests — but every plan reaches every model in the catalog, because credits pay for anything above that limit. Plans renew monthly through PayPal and cancel from your account; checkout on the plans page needs a free account first.
Build your setup once
Pick a model, create an API key, and attach any public or personal prompts, lorebooks, and plugins you want. Composite prepares those tools for each request while its context-management systems keep long conversations efficient. Your roleplay frontend continues using a familiar OpenAI-compatible connection.
Create a key
A free account issues one on the dashboard. The same key reaches the whole catalog, so you never manage a second one.
Paste the endpoint
Point your app at https://composite.lucidity.sh/v1 and pick a model. The docs list the exact field for each client.
Attach your tools
Bind prompts, lorebooks, presets and a memory layer to the key. They apply to every request from then on, from any frontend.
Works with the roleplay client you already use
Any frontend that accepts a custom OpenAI-compatible provider works: paste the endpoint, paste the key, pick a model. Each guide lists the exact fields for that app and the errors people hit most often.
JanitorAI proxy setup
Use Composite as a free JanitorAI proxy: the full chat-completions URL to paste, how the free daily tier works, and fixes for 429 and invalid-key errors.
SillyTavern setup
Connect through SillyTavern's Custom OpenAI-compatible Chat Completion source, including the base URL, key field, and model list refresh.
ChubAI setup
Point ChubAI at Composite as a custom API so character cards run on whichever model you choose.
Agnai setup
Configure Agnai's OpenAI-compatible preset to run its characters through Composite.
RisuAI setup
Add Composite as a custom OpenAI provider in RisuAI and keep your existing prompt setup.
Your own app
Point any OpenAI SDK at the endpoint, or use the Anthropic messages, Gemini, and OpenAI Responses formats directly, the same key works across all of them.
Bring your next story to life.
Create a free account, make a key, and connect the roleplay client you already use.
Create accountQuestions people ask before signing up
What is an OpenAI-compatible API?
An OpenAI-compatible API accepts the same request and response shape as
OpenAI's chat completions endpoint, so any client that can talk to OpenAI can talk to it by
changing two fields: the base URL and the API key. Composite serves that shape at
/v1/chat/completions, and also speaks the Anthropic /v1/messages,
Google Gemini, and OpenAI Responses formats, so the same account works whichever SDK you
already use.
Is there a free AI API for roleplay?
Yes. Composite runs a free daily tier of open-weight models, including Llama 3.3 70B, DeepSeek V4 Flash, GLM 5.2, Kimi K2.6, GPT-OSS 120B, Gemma 4, and Nemotron 3. Accounts are free and no credit card is required; connecting a Discord account unlocks the free tier.
Which AI model is best for roleplay?
There is no single answer, because the tradeoff is between prose quality, context length, and price. MiniMax M3 and GLM 5.2 are cheap, long-context defaults for everyday character chat; Claude Sonnet 5 and Claude Opus 5 write the most controlled long-form prose; DeepSeek V4 Flash and the free open-weight tier cost little or nothing for casual scenes. Every model has its own page with context window, pricing, and a roleplay fit note, and the best roleplay LLMs guide compares them side by side.
Can I use Claude, GPT, and Gemini through a single API key?
Yes. One Composite key reaches every model in the catalog, so switching between Claude, GPT, Gemini, DeepSeek, Kimi, Qwen, GLM, and open-weight models is a change to the model field rather than a new account, a new key, or a new billing relationship.
How much does an AI roleplay API cost?
It depends on the model and on how long your conversations get, because you pay per token rather than per message. Composite passes output tokens and fresh input through at catalog price, and bills the repeated part of a prompt below list price once the provider is serving it from cache, so a long chat history costs less to resend each turn. Free-tier models cost nothing, and monthly subscription plans cover a fixed daily allowance instead of per-request billing.
What happens when a roleplay outgrows the model's context window?
By default the oldest messages are trimmed so the request still fits, which is why long chats start forgetting early details. Composite's memory layers instead summarize old history, keep a durable record of characters, relationships, and open plot threads, and retrieve older passages that relate to the newest message, so a long-running story keeps its established facts.
Why should I use Composite over other cheap inference providers?
Every request goes through a paid, legitimate upstream provider, and the repeated part of a long prompt is billed below the list price once it is cached.
What is your upstream provider?
For non LucidityAI models, OpenRouter is the upstream provider. For LucidityAI models, Composite uses its own infrastructure.
Is there a free tier?
Accounts are free, and every account gets a daily allowance of free-model requests. No Discord account, credit card, or invite is needed.
Can Composite be trusted to not disappear?
Composite has been running for about a year. Your credits, keys and saved assets stay in your account, and you can export or delete them at any time.
Is this the same Composite as "Composite Seabase"?
Yes. "Composite Seabase" refers to Composite, the OpenAI-compatible AI platform for roleplay and apps.
Is Composite compatible with JanitorAI and SillyTavern?
Yes. Composite works with JanitorAI, SillyTavern, ChubAI, Agnai, RisuAI, and essentially any roleplay frontend that accepts an OpenAI-compatible endpoint.
Is there a free JanitorAI proxy?
Yes. Composite gives every Discord-verified account a daily allowance of free-tier models with no credit card required. See the JanitorAI proxy guide for setup.
Does Composite work with SillyTavern?
Yes. SillyTavern connects to Composite through its Custom OpenAI-compatible Chat Completion source. See the SillyTavern proxy guide for the exact fields to paste.
How do I get a JanitorAI API key?
Create a free Composite account, verify Discord to unlock the free daily model tier, then generate a key on the API Keys page. No credit card is required.