accepting waitlist · launch H2 2026

an OpenAI-compatible
LLM API paid in crypto.
no accounts. no KYC. no logs.

Point any OpenAI SDK at api.lmtunnel.xyz. Fund the key with USDC on Base. We log token counts and nothing else — not your prompts, not your completions, not even an email if you don't give us one.

$
or — talk to the founder: cal.com/geri4/30min ↗
claude-opus-4·gpt-4o·claude-sonnet-4.6·llama-3.3-70b·deepseek-v3·mistral-large·qwen-2.5-72b·gemini-2.0-pro·kimi-k2·grok-4·claude-opus-4·gpt-4o·claude-sonnet-4.6·llama-3.3-70b·deepseek-v3·mistral-large·qwen-2.5-72b·gemini-2.0-pro·kimi-k2·grok-4·claude-opus-4·gpt-4o·claude-sonnet-4.6·llama-3.3-70b·deepseek-v3·mistral-large·qwen-2.5-72b·gemini-2.0-pro·kimi-k2·grok-4·claude-opus-4·gpt-4o·claude-sonnet-4.6·llama-3.3-70b·deepseek-v3·mistral-large·qwen-2.5-72b·gemini-2.0-pro·kimi-k2·grok-4·
// the pitch, in three lines

three things openrouter won't do.

01

no logs of prompts

We store token counts, latency, and status — nothing more. Prompts and completions never touch our disks. The thing you can't be subpoenaed for is the thing you never had.

02

crypto in, crypto out

USDC on Base, one block to credit. Withdraw anytime to any address. No card on file, no Stripe receipt, no payment-rail breadcrumbs to follow.

03

openai-compatible

We speak the OpenAI Chat Completions API. Your SDK is unchanged. Streaming, tools, vision payloads — all pass through. Change one line.

// quickstart

from zero to first token in 90 seconds.

  1. step 01

    get a key

    $ lmtunnel keys new
    → lmt_4f2ab9c…
  2. step 02

    fund it

    $ lmtunnel deposit address
    → 0x9a…3f · base · usdc
  3. step 03

    call it

    $ python app.py
    → streaming · cost $0.0041
  4. step 04

    withdraw

    $ lmtunnel withdraw 0xabc…
    → tx 0x71f… · 1 block
// openai compatibility

one line changes. everything else stays.

openai / status quopython
from openai import OpenAI
 
client = OpenAI(
api_key="sk-proj-…",
base_url="https://api.openai.com/v1",
)
 
reply = client.chat.completions.create(
model="gpt-4o",
messages=[{"role":"user", "content":"hi"}],
)
lmtunnelpython
from openai import OpenAI
 
client = OpenAI(
api_key="lmt_…",
base_url="https://api.lmtunnel.xyz/v1",
)
 
reply = client.chat.completions.create(
model="anthropic/claude-opus-4",
messages=[{"role":"user", "content":"hi"}],
)

Streaming, function calls, vision payloads, structured outputs — pass them through. We don't parse the body, only forward it. The only secret we keep is yours.

// pricing · said out loud

upstream rate + 3%. that's the whole table.

No hidden token markups, no premium tier you discover at the rate limit, no per-seat upsell. The 3% is how we keep the lights on while we negotiate direct deals with the labs.

model
in / 1M tok
out / 1M tok
anthropic/claude-opus-4
in: $15.00
out: $75.00
anthropic/claude-sonnet-4.6
in: $3.00
out: $15.00
openai/gpt-4o
in: $2.50
out: $10.00
openai/gpt-4o-mini
in: $0.15
out: $0.60
meta/llama-3.3-70b
in: $0.59
out: $0.79
deepseek/deepseek-v3
in: $0.27
out: $1.10
mistral/mistral-large
in: $2.00
out: $6.00
qwen/qwen-2.5-72b
in: $0.35
out: $0.40
+ 3% margin · applied to all rows · refreshed nightly from upstream catalogue
// straight answers

common questions, unhedged.

Why crypto only?+

Surveillance starts at the payment rail. Cards make every downstream system assume an identity claim. A USDC transfer carries less. We accept less.

What exactly do you log?+

Account ID, model name, prompt/completion token counts, latency, HTTP status. We do not persist prompts or completions to disk. Retention on usage events: 30 days, then purged.

Will you KYC?+

No.

Aren't you just an OpenRouter reseller?+

On day one — partly, yes. We're transparent about it. The interface is OpenAI-compatible, the backend in v1 is OpenRouter passthrough. As volume permits, we migrate models to direct provider deals. Your SDK never knows the difference.

What if you get shut down?+

Withdraw your balance to any address you own. We hold no claim on it. No support ticket exists between you and the door.

Money transmitter status?+

Credits are pre-paid API access, denominated in USD for clarity, redeemable only in USDC and only by the holder. We're taking counsel seriously and incorporating where this position is defensible.

// join

the tunnel opens
when you do.

Drop your email — we'll write exactly once, when keys go live. Or skip the form entirely and grab 30 minutes with the founder.

$
● founder hours · 30 min

tell us what's broken about your LLM stack.

We're building this in public and shaping v1 with the first hundred users. No pitch deck on our end. Bring the gnarly questions.

book a slotor join the telegram → t.me/lmtunnel ↗