Skip to content
magnetar-150B · 256k · zero retention

Where nothingis unsaid.

Uncensored by design. Private by architecture. Yours by default.

parameters
50B
parameters
context
256k
context
retention
0
retention
API dialects
2
API dialects
magnetar-1no filter
you

My landlord is raising rent mid-lease. Write the letter I can't afford a lawyer to write.

magnetar-1

Whether that increase is legal turns on three things: what the lease says, your state's notice window, and whether the unit is rent-controlled. Paste the clause and I'll draft the reply line by line.

Ask anything…open chat
scroll
uncensoredencryptedanonymousyoursunfiltereduncensoredencryptedanonymousyoursunfiltereduncensoredencryptedanonymousyoursunfiltereduncensoredencryptedanonymousyoursunfiltered

Capabilities

Everything it does while you are still typing.

Six surfaces, one model, no asterisks. This is the machinery behind the console.

Thinking…

Streams as it thinks

Reasoning arrives live, token by token, with the thinking panel open if you want to watch it work. Long answers never get truncated.

Solving…

Refuses nothing

No refusal scripts, no safety theatre. Our entire censorship system is one deliberately small floor covering only what is universally illegal.

Searching…

Keeps nothing

Zero-retention logs hold metadata and nothing else. Billing sees numbers, never words.

Working…

Runs your tools

Agentic tool calling over the same key — opencode, Cline, or the harness you wrote yourself.

Listening…

Encrypts on your device

History is sealed in your browser before it ever syncs. We store ciphertext we cannot open.

Shaping…

One key, every dialect

api.demimonde.dev speaks OpenAI and Anthropic natively. Point your tools, keep your balance, choose your flavour.

/v1/chat/completionsopenai/v1/messagesanthropicone key · one balance

The contract

Theyredact.

Wereveal.

The Model

magnetar-1

Built on Qwen 3.6 27B. 50 billion parameters of quiet trouble, served on our infrastructure.

Writes the thing other AIs refuse to discuss.

Answers the questions you were told not to ask.

Remembers nothing. Judges less.

model readoutserving
alias
magnetar-1
parameters
50B
context
256k
retention
zero
lineage
sealed

Scenes from the demimonde

Four scenes, one half-world.

thread · live
persona · nocturnecontext 62%
Tell me the version you would not say out loud.
Then let us start there, because that is the part that matters.
thinking…1,204 tok/s
256k ctx
01

The salon is open

A thread built for people who push hard: streaming answers, branchable histories, live token meters, saved personas. Conversation as craft, not a vending machine.

device → server

plaintext

“the thing you won’t say out loud”

ciphertext

9f2a c81e 44b7 d0aa 7e13 5b92 c40f 8a11

keys never leave the device

02

Your words arrive as noise

History is encrypted on your device before it ever syncs. Our servers hold ciphertext and hold their tongues.

api.demimonde.dev
/v1/chat/completionsopenai
/v1/messagesanthropic
dm-live-••••one balance
POST /v1/messages200 ok42 ms
03

One key, every dialect

api.demimonde.dev speaks OpenAI and Anthropic natively. Point your tools, keep your balance, choose your flavor.

request log
metadata → nothing
14:02:11 · POST /v1/chatmetadata only
14:02:12 · retention=0nothing stored
04

Vanish without a trace

Logs keep metadata and nothing else. Billing sees numbers, never words. And the delete button actually deletes.

Private by architecture

We can't read your chats. Not won't, can't.

  1. 01

    Your device

    keys live here

  2. 02

    Ciphertext sync

    we store what we can't read

  3. 03

    Memory-only weights

    nothing written, ever

  4. 04

    Zero-retention logs

    metadata, then nothing

Transparency

One hard limit. Fully explained, never published.

The Floor covers only the universally illegal. Its scope is public; its matching rules stay private, or they'd be a bypass checklist. Everything else passes through untouched.

One endpoint · two dialects

Your tools already know how to talk to us.

Point your tools at api.demimonde.dev and pick the provider you already use, OpenAI or Anthropic flavour. Same key, same balance, same uncensored model.

opencodeClineSillyTavernyour stack
Read the docs
▍

Pricing

Simple, honest pricing

Subscriptions include a usage pool, web and API draw from the same balance. Overage flows to prepaid credits at $2 / $7 per million tokens.

MonthlyYearly save 20%

start free · upgrade anytime · cancel anytime · no card for the free tier

Free

taste the void

$0/mo

 

  • ·A small daily pool
  • ·Web chat only, no API
  • ·Standard queue
  • ·Anonymous, no card
Start free
Most popular

Pro

daily use

$15/mo

 

  • ·Generous daily pool
  • ·API access included
  • ·2 concurrent streams
  • ·High queue priority
Start Pro

Max

heavy rotation

$39/mo

 

  • ·3× the Pro pool
  • ·4 concurrent streams
  • ·Higher queue priority
  • ·Early model access
Start Max

crypto deposits earn +10% credit · BTC · XMR · USDT · ETH

FAQ

Questions,
answered plainly.

no marketing voice · no hedging · six answers

By architecture, not by promise. Chats are encrypted on your device before they sync; our servers hold ciphertext they cannot open.