Private AI: no identity, no retention, no training
Private AI means three things when the model is behind an API: no identity to open the account, no prompt or completion kept after the request, and nothing you send used to train anything. RelaySwitch runs 338 models under those three rules, on one key topped up in USDT, USDC, BTC, ETH, SOL or by card.
What private AI means
Three questions to ask any provider, and the answers here, from the privacy page.
No identity
What is asked for before you can call a model. Here: an email address and a top-up. No identity documents, no phone number, no business registration, for crypto and card alike.
No retention
What is kept after a request. Here: one usage record with the request id, model, token counts and charge. Request and response bodies are not written anywhere.
No training
Whether your prompts teach anyone’s model. Here: nothing you send is used for training, and the providers behind the catalog receive only the content of the calls they serve.
What is kept, and what is never kept
The whole list, from the privacy page.
Kept
- The email address the account was opened with
- Each top-up: amount, method and the payment’s own reference
- Each key: name, status, limits, timestamps
- One usage record per request: request id, model, token counts, the charge
Never kept
- Identity documents, a phone number, a name
- Request bodies: prompts, system prompts, documents
- Response bodies: completions
- Anything used to train a model
Redaction and per-key switches
What the model sees is yours to narrow. One header for a request, or a default on the key.
curl https://api.relayswitch.io/v1/chat/completions \ -H "Authorization: Bearer sk-relay-..." \ -H "x-uplink-redact: secrets_pii" \ -H "Content-Type: application/json" \ -d '{"model": "MODEL_ID", "messages": [{"role": "user", "content": "Debug this: AWS_SECRET=... failed for jane@example.com"}]}'# The model sees [[SECRET_1]] and a placeholder for the email.# x-uplink-redactions: 2 The answer carries the originals back.secrets replaces API keys, tokens and private keys with placeholders before the request is forwarded; secrets_pii also replaces emails, phone numbers, card numbers, US SSNs and IBANs. The answer gets the originals back. Set it per request with the header, or as the key's default in the console, so an agent that pastes a log line never sends the token in it.
Each key also carries its own allowed-model list, budget, rate limits and expiry, and keeps no bodies unless retain_bodies is switched on for it. Details on the files and redaction page.
The flagship models, under the same rules
USD per million tokens, read from the catalog at render. The developer's list price beside ours where one is on record.
| Model | Input | Output | List input | List output | Under list |
|---|---|---|---|---|---|
| Claude Opus 5.5claude-opus-5-5 | $2.80 | $10.00 | $4.00 | $20.00 | 50% |
| DeepSeek V4.1 Flash 1017deepseek-v4.1-flash-1017 | $0.25 | $1.20 | $0.30 | $1.20 | — |
| GLM 5.3 Flashglm-5.3-flash | $0.09 | $0.15 | $0.15 | $0.50 | 70% |
| GPT-6 Lunagpt-6-luna | $0.0625 | $0.20 | $0.10 | $0.50 | 60% |
| Qwen3.8 Maxqwen3.8-max | $0.80 | $2.40 | $2.00 | $6.00 | 60% |
| Minimax M3.1 Flashminimax-m3.1-flash | $0.105 | $0.42 | $0.15 | $0.60 | 30% |
| Gemini 3.8 Flashgemini-3.8-flash | $0.4925 | $2.25 | $0.75 | $3.75 | 40% |
| Grok 4.7grok-4.7 | $0.60 | $1.80 | $2.00 | $6.00 | 70% |
338 models in all on the Models page, the uncensored builds included. Every one is billed per token from the same prepaid balance.
Private AI: questions
What total privacy would mean, and what this API keeps.
An email, a top-up, a key
Prepaid from $10 in USDT, USDC, BTC, ETH, SOL or by card. No identity documents, no stored prompts, no subscription.