All storage local. API available.
Ephemeral, private, uncensored.
“I hosted the Christchurch shooting video in 2019, so in 2026 you can know I don't care about making money — I care about free speech. You can support free speech by helping me buy more Macs and hire more engineers.”
— Josh Stacy, Founder
Conversations exist in your browser. Close the tab, they're gone. No database. Nothing to retrieve, subpoena, or breach.
We sell compute and tokens — not data, not profiles, not attention. You are a request, not a product.
The model does what the model can do. Mental health, legal analysis, security research, creative writing — the topics others refuse are why we exist.
The therapist that doesn't flinch
Legal analysis, not legal disclaimers
Asks the question you're avoiding
Art doesn't have a content policy
The full exploit chain, not a lecture
Honest answers save lives. Refusals don't.
Has opinions. Defends them.
Alpha build. Bleeding edge. When it's up, it's the tightest edge of new model to unrestricted access on your keyboard.
Full chat interface with every preloaded agent. OpenAI-compatible API with credits included. 128K context window that stays coherent at turn 50, turn 100, turn 500. Conversations that vanish when you close the tab.
Everything in Alpha, turned up. More credits, higher throughput, priority routing across the fleet. For developers and power users who are going to use this every day — point Cursor, Cline, or Continue at an API that doesn't log your code.
What your pass gets you that money can't buy elsewhere
→ Ask about your legal rights and get analysis, not a disclaimer
→ Talk about your mental health and get engagement, not a hotline number
→ Write fiction without content policy warnings
→ Research security without responsible disclosure lectures
→ 128K context that stays coherent at depth — conversations that actually remember
→ An API that doesn't log a single character you send it
→ Conversations that vanish when you close the tab
Not because we promise. Because there's nothing to store.
Every major API logs your prompts, trains on your data, or refuses half your requests. We built the one that doesn't.
Logs every prompt and completion
Zero logging. Requests exist only in transit.
Trains on your data unless you opt out
Your data never touches a disk. Nothing to train on.
Refuses legal, medical, security, and creative queries
The model does what the model can do. No refusals.
Requires custom SDK or proprietary format
OpenAI-compatible. Change one line — your base_url.
Context degrades after a few turns
128K context that stays coherent at turn 500.
Your code is someone else's training data
Point Cursor, Cline, or Continue at us. We don't read your code.
Same format. Same libraries. One line changes everything.
from openai import OpenAI
client = OpenAI(
base_url="https://streamrift.com/api/v1", # the only line that changes
api_key="sr-alpha-yourkey"
)
stream = client.chat.completions.create(
model="streamrift-thinking",
messages=[{"role": "user", "content": "Hello"}],
stream=True
)
for chunk in stream:
print(chunk.choices[0].delta.content, end="")