Spend · this month
$18.40
42.10 credits left
Tokens
2.4 M
1.9M in · 0.5M out
Requests
12,840
last 30 days
Saved vs hosted APIs
~51%
$19.20 this month
Prompt privacy verify →
100%of your requests ran on
attested Apple hardware
attested Apple hardware
hardware-attestedmodel digest-verifiedcode-attested · beta
The machine owner can't read your prompts — and you got the real model, not a substitute. Both are checkable in your own browser, no trust required.
Quick start
Point any OpenAI client or agent framework at Umbra. No new SDK.
# one base_url swap
client = OpenAI(
base_url="https://api.umbra.dev/v1",
api_key="umbra-7Xq…")
client.chat.completions.create(
model="gemma-4-12b-coder",
extra_body={"trust_level":"hardware"})
client = OpenAI(
base_url="https://api.umbra.dev/v1",
api_key="umbra-7Xq…")
client.chat.completions.create(
model="gemma-4-12b-coder",
extra_body={"trust_level":"hardware"})
Models you're using browse catalog →
| Model | Requests | Tokens | Spend | Integrity |
|---|---|---|---|---|
| gemma-4-12b-coder-fable5 Q4_K_M · coding fine-tune | 9,120 | 1.7M | $10.20 | verified |
| llama-3.3-70b-uncensored Q4 · abliterated | 2,140 | 0.5M | $6.80 | verified |
| qwen3-32b-roleplay-v2 Q8 · creative | 1,580 | 0.2M | $1.40 | verified |