Private AI chat — no account, no plaintext on a server
The short answer
ShadowTalk AI is a free AI chat where you can start as a guest with no account, send cloud requests browser-to-provider with your own API key (BYOK), run the model fully on your device, and switch on end-to-end encryption (AES-256-GCM) for stored chats in Private Mode. You choose the privacy tier per conversation — including the tier where no server ever sees plaintext.
"Private" is the most abused word in AI. This page defines exactly what ShadowTalk does at each tier, where your keys live, what the server can and cannot see, and where the honest limits are — so you can decide instead of trusting a slogan.
What "private AI chat" should mean
Six checks any private AI chat should pass. ShadowTalk passes them at some tier — and tells you exactly which:
- No account required to start — guests can chat immediately, including with a local model
- Your keys stay yours — BYOK keys live in your browser and are sent directly to the provider you chose
- No training on your conversations — ShadowTalk never trains foundation models on your chats
- An on-device option — inference can run in your browser with zero network egress
- Encryption at rest you control — Private Mode encrypts chats end-to-end before storage
- Honest server story — documented, not hidden: what defaults store, what free shared mode transits
Three privacy tiers you pick per chat
ShadowTalk's privacy is a dial, not a promise:
- Tier 1 — Guest + on-device: no account, model runs in your browser, zero network egress. Nothing to intercept because nothing is sent.
- Tier 2 — BYOK direct: paste an API key from any of 16+ providers (OpenAI, Anthropic, Google, Groq, DeepSeek, OpenRouter and more). The key is stored in your browser only, and requests go straight from your browser to the provider — including Anthropic's direct browser access header — with no ShadowTalk proxy in the path.
- Tier 3 — Private Mode: stored chats, titles and metadata are encrypted end-to-end with AES-256-GCM. Keys are derived on your device from your passphrase, held as non-extractable browser keys, and kept memory-only. Turning Private Mode on mid-conversation retroactively encrypts the whole history. Storage follows an encrypt-or-null contract: if encryption fails, nothing is written rather than silently falling back to plaintext.
How that compares to the privacy-first names you know
Proton Lumo protects chat history with zero-access encryption — but inference is cloud-only (its published security model describes the server decrypting your prompt to process it), it needs an account and it needs the internet. Secret Llama and web-llm-chat run fully in your browser with no server — but they have no sync, no backup and no upgrade path to cloud models. ShadowTalk is the only product that combines on-device inference, end-to-end encrypted storage, cross-device sync and BYOK direct-to-provider in one browser tab — free.
Honest limits (read this part)
Privacy claims die in the fine print, so here is ours. Default chats — when Private Mode is off — are stored server-side in Firestore. The free shared mode routes cloud requests through ShadowTalk's server, which necessarily sees that mode's prompts. The service includes analytics events. Server-side data has admin read access in the database rules. If you need zero server exposure, that's exactly what Tiers 1-3 are for: on-device mode, BYOK direct requests, and Private Mode encryption. Most products hide this paragraph; shipping it is the point.
Frequently asked questions
Keep reading
Free to use — no account, no card.
Works offline. Syncs when you're back. Your keys, your models, your chats.
Start private AI chat — free