Private AI.
No logs.
No names.
Open-weight models without added content filters — in the browser or through an OpenAI-compatible API. Your account is a random key. Personal history stays in your browser; shared team history is stored encrypted.
Our server and the inference provider process your messages while answering. This is not end-to-end encrypted inference. Privacy details.
- 0 B
- plaintext prompts stored
- 120
- bit account key
- CRYPTO
- USDT · USDC
// how it works
Privacy by architecture
Personal history stays local, team storage is encrypted, and inference still requires plaintext processing.
A key, not an identity
Your browser completes a self-hosted CAPTCHA and gets a random 120-bit account key. That's the only credential — we keep just a keyed hash of it.
History stays with you
Personal conversations live in your browser. Publish selected chats to an encrypted team space, or fork a private copy of shared history. The server relays plaintext only for inference.
No request logs
No plaintext prompts, answers, IP addresses or user agents are logged. Usage accounting keeps a daily credit total normally for 35 days; shared-history storage holds ciphertext and update metadata.
Crypto, no processor
USDT / USDC sent straight to our wallet. No card processor and no buyer details requested.
Honest about limits
Our server and third-party inference infrastructure process plaintext messages while answering. Providers are selected for no-retention policies; we cannot technically audit their compliance. The privacy policy spells it out.
// the difference
What they keep. What we keep.
| Typical AI app | Cyberdeck | |
|---|---|---|
| Sign-up | ✕Email, phone, often a card | ✓A random 120-bit key |
| Your conversations | ✕Stored on their servers | ✓Personal: local · Team: encrypted |
| Request logs | ✕IP, device, timestamps | ✓None written |
| Payments | ✕Card processor, your name | ✓Crypto only · no card processor |
| Content filters | ✕Added on top of the model | ✓None added to open weights |
| Leaving | ✕Support ticket, maybe | ✓Delete your account in the app |
// the index
Models
Every current model is available on Free and paid plans. API access depends on your plan; context is capped by both the model and plan.
| Type | Name | Model context limit | Credits / token | Shared provider slots | Available from |
|---|---|---|---|---|---|
| LLM | Specter Code and reasoning: agents, debugging, analysis, long contexts. specter | 200K | 1 out / 0.175 in | 32 | Free+ |
| LLM | Quill Creative writing: stories, posts, copy, editing, any language. quill | 200K | 1.25 out / 0.4 in | 16 | Free+ |
// pricing
Prepaid. No subscription.
Pay once for 30-day months. No automatic renewal; when time ends you return to Free. All plans use the same models and privacy protections. Paid plans add credits, context, priority and the features listed below.
Free
- 50K credits / UTC day
- 200K credits / rolling 7 UTC days
- 16K-token context
- 1 simultaneous generation
- Chat only · no API
- All models, standard queue
Lite
- 5M credits / 30-day cycle
- 32K-token context
- 1 simultaneous generation
- Chat only · no API
- All models, priority queue
Plus
API- 40M credits / 30-day cycle
- 32K-token context
- 2 simultaneous generations
- Chat + OpenAI-compatible API
- All models, priority queue
- Eligible Specter cached input −12.5%
Pro
API- 150M credits / 30-day cycle
- 64K-token context
- 3 simultaneous generations
- Chat + OpenAI-compatible API
- All models, priority queue
- Rollover: up to one extra cycle
- Eligible Specter cached input −25%
Max
API- 350M credits / 30-day cycle
- 128K-token context
- 4 simultaneous generations
- Chat + OpenAI-compatible API
- All models, priority queue
- Rollover: up to one extra cycle
- Eligible Specter cached input −37.5%
Ultra
API- 650M credits / 30-day cycle
- 200K-token context
- 6 simultaneous generations
- Chat + OpenAI-compatible API
- All models, priority queue
- Rollover: up to one extra cycle
- Eligible Specter cached input −50%
What credits cover. Each request spends credits on input — including conversation history sent again — and generated output, at the model rates above. One credit equals one generated Specter token; input is charged too, so credits are not a fixed number of messages or universal tokens. Cache discounts apply only to provider-reported cached input on eligible models and plans, never to all input or output. Cache hits are not guaranteed.
Optional web search. Ask one question; the model plans the searches. Free includes one query for 25,000 credits. Paid plans include up to 3 related queries for one 50,000-credit fee, even when fewer are needed. Planning is included; answer and evidence tokens are charged normally. The fee is retained after the first successful lookup, even if it finds no results or the reply is stopped. Packs do not unlock the paid search cap.
Credit packs. Ran out mid-project? Top up any plan, Free included: 12.5M credits for $10 · 50M credits for $40 · 200M credits for $160. Spent after your own plan and available team credits; no expiry while your account exists. Packs do not extend plan time or unlock API, larger context, concurrency or team features.
Teams. Pro+ accounts can pool their plans: everyone keeps their own plan and API keys, and when someone runs out they draw from teammates' unused credits. Invite by one-time link, up to 20 members, optional per-member caps. Share encrypted chats, branch from any message, or keep a private fork in your browser.
Plan windows, extensions and rollover
Free must fit both its UTC-day limit and its rolling 7-UTC-day limit. Paid credits are granted per 30-day cycle; buying several months does not grant every cycle's credits at once.
Extending the same active plan adds time, not an immediate refill. Switching converts remaining time by value and starts a new cycle; usage from the activation day still counts. Existing rollover is forfeited when you switch or let a plan expire.
On rollover plans, unused credits carry into the next cycle, capped at one base cycle's extra allowance, while you keep that plan active. Other plans do not carry unused cycle credits.
Context means instructions, conversation history and generated output together, limited by both model and plan. Concurrency is simultaneous generations across chat and API. Queue priority does not guarantee response speed.
Crypto payments are final and network fees are yours. Review the exact token, network, amount and payment window at checkout. Payment terms.
// faq
Questions
I lost my account key. Can you recover it?
No. We only store a keyed hash of the key, so nobody — including us — can look it up. Save it in a password manager when you create the account.
What exactly do you store?
A hash of your account key, your plan and expiry, API key hashes, daily credit totals (normally 35 days) and payment records (30 days after final status). Unfinished credit reservations can retain related totals longer. Teams add membership and optional encrypted snapshots with update metadata. No plaintext prompts, answers, IPs or user agents are logged. The privacy policy covers retention, backups and the full list.
Who can see my messages?
Personal history stays in your browser. Shared team history is encrypted before storage and readable by teammates with its key. During inference, our server and the model provider process plaintext messages. We do not attach your account key or IP to inference requests. Providers are selected for their non-retention policies, but we cannot technically audit them. Do not send anything you could not accept an infrastructure provider seeing.
Which coins do you accept?
USDT on Tron (TRC20), USDT on BNB Chain (BEP20), USDC on BNB Chain (BEP20), USDT on Ethereum (ERC20), USDC on Ethereum (ERC20), USDT on Polygon, USDC on Polygon. Public-chain transfers are pseudonymous, not anonymous.
Refunds?
Plan time is prepaid and crypto payments are final. Switching plans converts your remaining time by value. See the terms.
Why is a reply taking a while?
A request may wait for an available slot or for the inference provider to respond. Paid plans get queue priority, not a guaranteed response time. The chat shows the status it knows; a delay alone does not tell us its cause.
// set sail