e2ee chat · abliterated models · OpenAI-compatible
Chat that is encrypted in your browser and opened only inside verified hardware. Open-weight models with the refusals taken out. One API, and every model labelled with exactly who can read your prompt.
No account or card needed to chat.
on the wire 8f3a·c19e·77d0·b2e4·0a5f…
attestation verified · enclave key pinned
readable by you
Every model ID starts with its privacy mode, and every response carries an x-garlic-privacy header. You choose the trade-off; Garlic never makes it silently.
Readable by
Only you
Encrypted in your browser, decrypted only inside attested hardware. Garlic's servers carry ciphertext.
Browser chat
Readable by
You and Garlic's server, in transit
Garlic's server verifies the enclave and encrypts to it, so the model host's operators cannot read the prompt.
API, attested models
Readable by
The GPU host
Plain TLS to a GPU host. Labelled as such on every response, never dressed up as private.
API, including abliterated models
The strongest privacy you can get from a hosted model: the prompt is sealed before it leaves the tab, and the hardware proves what it is before it gets a key.
Open-weight models with the refusal direction removed from the weights themselves. Not a jailbreak prompt that stops working next week, and not a lecture before the answer.
Runs in passthrough mode: the GPU host can read prompts, and Garlic retains requests encrypted for 30 days for abuse review. Use e2ee chat for anything sensitive.
Your message is locked in the browser before it leaves.
Encrypted locally
Garlic verifies the secure environment before anything is sent.
Attestation verified
Your prompt is decrypted only inside protected hardware, and the response returns encrypted.
Private session
Encrypted response
Garlic cannot read your encrypted chat content.
Chat completions, streaming, and usage reporting work as your SDK expects. Pick attested models for sensitive work and abliterated ones for everything a mainstream API refuses, from the same key.
Read the API referencemain.py
from openai import OpenAI client = OpenAI( base_url="https://garlic.ai/v1", api_key=os.environ["GARLIC_API_KEY"], ) reply = client.chat.completions.create( model="passthrough-runpod/qwen-abliterated", messages=[{"role": "user", "content": prompt}], )
Prepaid credits from $5 that never expire, or a membership with a monthly grant: Pro $20/mo, Power $60/mo, Max $200/mo. Browser chat is free to try.
Start a private chat without an account, or get an API key for the uncensored models.