Default Workspace/Home
Gateway overview, live routing playground & telemetrySentioGate is an AI LLM router: change one base URL and reach OpenAI, Groq, DeepSeek, Mistral, xAI, Claude and Gemini through OpenRouter, or your own vLLM/Ollama server, with the provider keys you already have. Your providers bill you directly; SentioGate adds nothing per token.
Change one line (base_url): every request goes to a capable model at a good price, falls back automatically when a model fails, and long conversations cost fewer tokens.
client = OpenAI( # Direct to single provider endpoint api_key="sk-proj-single_openai_key")res = client.chat.completions.create( model="gpt-4o", # Overpays on routine tasks messages=long_chat_history)client = OpenAI( base_url="https://api.sentiogate.dev/v1", api_key="ar_live_your_gateway_key",)res = client.chat.completions.create( model="auto", # Picks a capable model, falls back automatically messages=long_chat_history)Pick a sample workload or type your own to see how SentioGate classifies the task and picks a model.
| Metric | Before: direct provider callopenai/gpt-4o | After: SentioGate (model="auto")groq/llama-3.3-70b-versatile |
|---|---|---|
| Estimated cost | $0.005000 | $0.000030 (Lower cost) |
| Latency (TTFT) | Depends on one provider | Measured on a live run |
| Outage protection | Single point of failure (a busy or down provider drops the request) | Active circuit breaker + automatic model fallback |
| Context handling | Uncompressed raw tokens | Whitespace cleaned + edge cache |
* Cost estimates reflect input prompt and generated output tokens calculated at standard provider list prices. Actual savings depend on request length, cache hit rates, and the selected routing strategy (cost vs. speed).
Calibrated sample. Select “Route request” for a live run.
{
"entities": [
{"name": "Apple", "type": "Organization"},
{"name": "M4 Max", "type": "Product"},
{"name": "Cupertino", "type": "Location"}
],
"date": "today"
}Change the base URL and key. If a model errors or rate-limits, SentioGate automatically falls back to another model.
from openai import OpenAI
# 3-line drop-in: point directly to SentioGate router
client = OpenAI(
base_url="https://api.sentiogate.dev/v1",
api_key="ar_live_your_gateway_key", # or sg_live_...
)
response = client.chat.completions.create(
model="auto", # SentioGate picks the cheapest model that fits the task
messages=[{"role": "user", "content": "Explain vector embeddings in 10 words"}],
)
print(response.choices[0].message.content)Open-source SDK (Apache-2.0): sentiogate-python on PyPI. The SentioGate service itself is proprietary.
Measured from requests routed through SentioGate in the last 24h. Auto routing uses it as a secondary factor after price.
What is built and running today: one API, plus the controls, analytics and admin tools to run it.
Send model "auto" and each request goes to the cheapest capable model across your keys. Or pin any provider/model.
Automatically falls back to another model if one fails. Models that are struggling are skipped.
Trims long conversations so you spend fewer tokens. Each response shows what was saved.
Spots an agent repeating the same steps and stops the request before it burns more tokens. You can switch it off.
Prefer cost, quality or speed, hint the task type, and switch caching, trimming, loop stop or the dangerous tool-call flag on or off per request.
Requests, spend at your providers' prices, cache savings, latency, success rate and recent routing decisions. Optional trace export to your own Langfuse.
Create and revoke gateway keys, keep up to 5 keys per provider in fallback order, turn models on or off, and set spend limits that pause a key.
Add any OpenAI-compatible HTTPS endpoint, such as vLLM or Ollama, with custom headers.
Keys are encrypted at rest and never shown back in full. Providers bill you directly; SentioGate charges only for the plan.
Connect Claude Code, Cursor, Cline or Hermes via npx -y @sentiogate/mcp or POST /v1/mcp to audit spend, check guardrails and route sub-tasks.
Sourced guides on BYOK LLM routing, provider rate limits, and gateway architecture.
Prototype with 600 requests every 5 hours on the free community pool (no credit card), or route across free-tier keys.
Routers, gateways and direct APIs compared with sourced fees and free tiers, and when to stay on OpenRouter.
Fallback when OpenAI, Gemini or Groq is busy or down, OpenRouter/LiteLLM/Portkey/Helicone alternatives, and cost controls.