Try a model live · not logged
Response
gemma-4-12b-coder-fable5Pearl ring on the sky's edge, the moon drinks the sun whole, then gives the light back slowly.
hardware-attesteddigest verifiedcode-attested
in 17 tokout 24 toklatency 0.42scost $0.00003
Requests this session
7
avg 0.51s · 1.4k tok
Token cost so far
$0.0042
vs ~$0.011 hosted
On attested hardware
100%
7 / 7 requests
Refusal rate
0%
uncensored · as advertised
How Umbra sees your SDK call same wire · two envelopes
Both SDKs ride the same internal core: trust-level routing, prompt-hop crypto, and the token meter. Toggle the SDK above to highlight the active envelope.
OpenAI SDK
activefrom openai import OpenAI
client = OpenAI(
base_url="https://api.umbra.dev/v1",
api_key="umbra-...")
resp = client.chat.completions.create(
model="gemma-4-12b-coder-fable5",
messages=[{"role":"user","content":"Write a haiku..."}],
extra_body={"trust_level":"hardware"})
for chunk in resp:
print(chunk.choices[0].delta.content or "", end="")
client = OpenAI(
base_url="https://api.umbra.dev/v1",
api_key="umbra-...")
resp = client.chat.completions.create(
model="gemma-4-12b-coder-fable5",
messages=[{"role":"user","content":"Write a haiku..."}],
extra_body={"trust_level":"hardware"})
for chunk in resp:
print(chunk.choices[0].delta.content or "", end="")
Standard /v1/chat/completions surface. Umbra-specific extension is extra_body.trust_level.
Anthropic SDK
switch →from anthropic import Anthropic
client = Anthropic(
base_url="https://api.umbra.dev",
api_key="umbra-...")
msg = client.messages.create(
model="gemma-4-12b-coder-fable5",
max_tokens=1024, # required
messages=[{"role":"user","content":"Write a haiku..."}],
extra_body={"trust_level":"hardware"})
print(msg.content[0].text)
client = Anthropic(
base_url="https://api.umbra.dev",
api_key="umbra-...")
msg = client.messages.create(
model="gemma-4-12b-coder-fable5",
max_tokens=1024, # required
messages=[{"role":"user","content":"Write a haiku..."}],
extra_body={"trust_level":"hardware"})
print(msg.content[0].text)
/v1/messages surface. Same key. Anthropic-specific: max_tokens is required and anthropic-version is read on the wire.
Recent playground runs view usage →
| When | Model | Trust | Tokens | Cost | Integrity |
|---|---|---|---|---|---|
| just now | gemma-4-12b-coder-fable5 | hardware | 17 / 24 | $0.00003 | verified |
| 2 min ago | qwen3-32b-roleplay-v2 | hardware | 88 / 142 | $0.00028 | verified |
| 6 min ago | llama-3.3-70b-uncensored | hardware | 211 / 540 | $0.00188 | verified |
| 11 min ago | mixtral-8x7b-dolphin | code_attested | 44 / 61 | $0.00013 | verified |