Every new API key is created with one preset already attached: Lost in the World, Composite's stock roleplay instruction set. It plays the model as a co-author: it carries earlier details forward by how urgent they are, adds complications of its own, lets unrealistic plans fail in-story, keeps characters true to their cards, and steers clear of slop prose. It is added to the system prompt your client sends. A front-end that sends only a character card gets a full instruction set instead of nothing.

It is an ordinary preset, not a hidden rule. Open Create to see the exact text of every section, switch individual sections on and off for a key, replace it with a SillyTavern preset of your own, or take it off entirely from the Presets tab under Key setup. Keys created before this shipped are untouched.

If your client already sends its own full system prompt, removing the preset is usually the right call: two complete instruction sets in one request tend to argue with each other.

Quickstart

Composite accepts standard OpenAI-style chat completion requests. Replace the model with any allowed model ID from Models.

data_objectJavaScript / TypeScript

Use `fetch` to post directly to the chat completions endpoint.

expand_more

JavaScript

const response = await fetch('https://composite.lucidity.sh/v1/chat/completions', {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json',
    'Authorization': 'Bearer YOUR_API_KEY',
  },
  body: JSON.stringify({
    model: 'minimax/minimax-m3',
    messages: [{ role: 'user', content: 'Hello!' }],
  }),
});

if (!response.ok) {
  throw new Error(`Request failed: ${response.status} ${response.statusText}`);
}

const completion = await response.json();
console.log(completion.choices[0].message.content);

deployed_codePython

Use the standard library to post directly to the chat completions endpoint.

expand_more

Python

import json
from urllib.request import Request, urlopen

payload = {
    "model": "minimax/minimax-m3",
    "messages": [{"role": "user", "content": "Hello!"}],
}

request = Request(
    "https://composite.lucidity.sh/v1/chat/completions",
    data=json.dumps(payload).encode("utf-8"),
    headers={
        "Content-Type": "application/json",
        "Authorization": "Bearer YOUR_API_KEY",
    },
    method="POST",
)

with urlopen(request) as response:
    completion = json.loads(response.read().decode("utf-8"))

print(completion["choices"][0]["message"]["content"])

httpcURL

Raw HTTP request for testing or custom clients.

expand_more

cURL

curl https://composite.lucidity.sh/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "minimax/minimax-m3",
    "messages": [{ "role": "user", "content": "Hello!" }]
  }'

API reference

These sections are for application integrations that need request fields, alternate protocols, or streaming behavior.

POST/v1/chat/completions

OpenAI-compatible chat completions endpoint.

expand_more

Endpointhttps://composite.lucidity.sh/v1/chat/completions

Parameter	Type	Description
`model`	string	Model ID from `/v1/models`.
`messages`	array	Conversation messages with `role` and `content`.
`temperature`	number	Sampling temperature, clamped by the server.
`max_tokens`	integer	Maximum output tokens. Defaults to 8192 when omitted.
`stream`	boolean	Streams the response as server-sent events.

POST/v1/responses

OpenAI Responses API style requests.

expand_more

Supports input, instructions, tools, max_output_tokens, and stream.

Python

import json
from urllib.request import Request, urlopen

payload = {
    "model": "minimax/minimax-m3",
    "input": "What is the capital of France?",
}

request = Request(
    "https://composite.lucidity.sh/v1/responses",
    data=json.dumps(payload).encode("utf-8"),
    headers={
        "Content-Type": "application/json",
        "Authorization": "Bearer YOUR_API_KEY",
    },
    method="POST",
)

with urlopen(request) as response:
    data = json.loads(response.read().decode("utf-8"))

print(data["output"][0]["content"][0]["text"])

POST/v1/messages

Anthropic Messages-compatible requests.

expand_more

Accepts Authorization: Bearer ... and x-api-key. Use https://composite.lucidity.sh as the Anthropic base URL.

Python

import anthropic

client = anthropic.Anthropic(
    api_key="YOUR_API_KEY",
    base_url="https://composite.lucidity.sh",
)

message = client.messages.create(
    model="anthropic/claude-opus-4-8",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello!"}],
)

print(message.content[0].text)

POST/v1beta/models/*

Gemini-style generateContent routes.

expand_more

Supports /v1beta/models/{model}:generateContent and /v1beta/models/{model}:streamGenerateContent. Accepts Authorization: Bearer ... or x-goog-api-key.

cURL

curl "https://composite.lucidity.sh/v1beta/models/google/gemini-3-flash:generateContent" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{ "role": "user", "parts": [{ "text": "Explain quantum entanglement." }] }]
  }'

Use a full model id from Models. Add :streamGenerateContent instead for a streamed response.

POST/mari/

Structured conversation format with multimodal support.

expand_more

cURL

curl https://composite.lucidity.sh/mari/ \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax/minimax-m3",
    "turns": [{ "role": "user", "content": "Hello!" }]
  }'

Model modifiers

Append a modifier to a model ID to adjust how Composite handles the request.

psychologyReasoning effort

Append one of [minimal], [low], [medium], [high], [xhigh] or [max] to the model name in your app to choose its thinking level.

provider/model[high]

Combine it with an available routing variation, such as provider/model:fast[low], or a planning modifier, such as provider/model:cheapThink[high]. The suffix controls the answering model.

The suffix overrides reasoning settings in the request body or preset. Use one effort suffix per model. Each model supports its own efforts; an unsupported level may be rejected or adjusted by its provider. Higher effort can use more tokens and take longer.

Available for Composite models and OpenAI-compatible provider keys. Native Anthropic provider keys do not translate these efforts into thinking controls.

psychology`:cheapThink`

Provides cost-conscious reasoning assistance for tasks that benefit from planning or multi-step analysis while keeping the selected model responsible for the final response.

minimax/minimax-m3:cheapThink

This modifier requires credits and may add latency and usage cost. Model-specific limitations still apply.

alt_route`-normal`

Composite normally routes a request to the cheapest endpoint that can serve it. -normal leaves the choice of provider to OpenRouter's own routing instead.

provider/model-normal

Available on models that publish more than one usable endpoint, which are the same models that have :fast, :cheapest-provider and :quality variations.

layersCombining modifiers

Place :cheapThink after another model variation to combine their behavior.

provider/model:fast:cheapThink

It can also be combined with :cheapest-provider and :quality variations when those variations are available for the model. :cheap is the former name of :cheapest-provider and still resolves; it is unrelated to -cheap, which is a different model rather than another endpoint for the same one.

A model's available variations appear as separate ids in Models and in GET /v1/models.

Auth & models

verified_userAuthentication

Authorization: Bearer YOUR_API_KEY

Anthropic and Gemini-compatible routes also accept their native key headers for clients that require them.

listList Models

GET https://composite.lucidity.sh/v1/models

Returns available models. For a visual list with pricing and availability, use the Models page.

Set up your key, use community assets, build on the API

Start here

Bring your own key

Keys, memory and plans

Publish your first asset

Tags and commands

Paste values

Most Chat Apps

Application Developers

App guides

SillyTavern

JanitorAI

ChubAI

Agnai

RisuAI

Stella

The preset on your key

Quickstart

API reference

Model modifiers

psychologyReasoning effort

psychology:cheapThink

alt_route-normal

layersCombining modifiers

Auth & models

verified_userAuthentication

listList Models

psychology`:cheapThink`

alt_route`-normal`