Your own AI, your rules.

Name it, pick a base model, write how it behaves. Use it in the chat and in any app that takes a model id. No training, no hosting.

Make one

Your own model, three ways

What each way asks of you.

A saved custom model here, against fine-tuning and running a model yourself
You needCustom model here
Start here
A saved preset on your account		Fine-tuningTraining a model on your dataRun it yourselfA model on your own machine
Training data or a GPUCustom model hereNoFine-tuningYesRun it yourselfYes
Somewhere to host itCustom model hereNoFine-tuningYesRun it yourselfYes
Time to the first replyCustom model hereMinutesFine-tuningHours or daysRun it yourselfHours
Personality and instructions built inCustom model hereYesFine-tuningYesRun it yourselfPer request
Backups when the model is busyCustom model hereUp to fiveFine-tuningNoRun it yourselfNo
Change it laterCustom model hereEdit and save, every version keptFine-tuningTrain againRun it yourselfReconfigure
PriceCustom model hereThe base model’s rate, nothing addedFine-tuningTraining plus hostingRun it yourselfHardware and power
Make one

What a custom model is here

A custom model is a model id you own. Behind the name sits a base model from the catalog, the instructions it always follows, the settings it starts with and the models that answer if the first one is busy. Save it once; from then on my-bot is a model like any other on your account, in the console and in any app that takes a model id.

  • A name: 1 to 64 characters, lower-case letters, digits, dots, dashes and underscores. A name cannot take over a catalog model’s id or alias.
  • A base model: any of the 152 chat models on the site, or an alias such as auto.
  • Up to five backups, tried in order when the base model does not answer.
  • Instructions of up to 32 KB that go in front of, or replace, what the app sends.
  • Defaults for temperature, top_p, max_tokens, stop and the rest. A value the app sends wins.
  • Optional tools and a description. Up to 100 custom models per account, every change kept as a version.

Make one

  1. Open Custom models in the console and choose a name.
  2. Pick the base model, and the backups if you want them.
  3. Write the instructions: who it is, how it talks, what it always does and never does. Set the defaults. Save.

The same thing by API is one request: PUT https://api.unrestrictable.ai/v1/presets/my-bot with an admin key or the console session. Everyday API keys read and use custom models and cannot change them.

Use it

Put the name where the model id goes. my-bot, @my-bot and @preset/my-bot all reach the same saved model. The OpenAI chat and Responses formats and the Anthropic Messages format all take it, and GET https://api.unrestrictable.ai/v1/models lists your custom models with the catalog, so a model picker in any app shows them.

main.py
1from openai import OpenAI
2
3client = OpenAI(base_url="https://api.unrestrictable.ai/v1", api_key="sk-unr-live-…")
4reply = client.chat.completions.create(
5 model="my-bot", # your saved model, by name
6 messages=[{"role": "user", "content": "Hello"}],
7)
8print(reply.choices[0].message.content)

In the chat, Settings holds a system prompt for one conversation; a custom model carries its instructions into every conversation and every app on the same prepaid key.

What it costs

A reply from my-bot is billed as a reply from its base model, at the live price per token on the models page, from the same prepaid balance. Nothing is added for the custom model. Top up from $10 with crypto; no plan, no ID, no phone.

Questions.

Yes. A custom model here is a saved preset: a name, a base model from the 152 chat models on the site, the instructions it always follows, its default settings and up to five backup models. Save it once and call it by name. Nothing is trained and nothing is hosted by you.

You can make one that talks like you: write how you speak, what you know and what you would say in the instructions, pick a base model, and every reply follows it. It does not learn from your messages and it does not become a copy of you; it is a base model following your instructions.

Saving a model costs nothing. A reply costs the base model's live price per token, taken from a prepaid balance you top up from $10 with crypto. There is no fee for the custom model itself.

It answers the way its base model does. Pick a base model from the catalog that is tuned without the refusal layer and your model inherits that; the acceptable use policy is the one filter that stays.

In the console, and in any app that takes a model id and a base URL: set the base URL to ours, your key, and the model name you saved. The OpenAI chat and Responses formats and the Anthropic Messages format all work, and the model list shows your custom models so pickers find them.

The backups you listed are tried in order, up to five. Your instructions and settings travel with the request.

Yes. Edit the instructions, the base model, the backups or the defaults; every change is saved as a new version and the name stays the same, so nothing that calls it has to change.

The preset, including its instructions, is saved on our side so the gateway can add it to each request. Your chats are not: prompts and replies pass through and are not stored, and we keep one usage record per request for billing.

Keep reading

One name. Your instructions, every time.

Save a model in the console and call it from any app. Top up from $10 with crypto.

See the base modelsMake one
Join on Telegram