One OpenAI-compatible endpoint & Model Context Protocol (MCP) Control Plane for the models you already pay for.
Routing by cost, quality or latency with Bring-Your-Own-Key (BYOK) and 0% token markup. SentioGate automatically falls back to another model if one fails.
Client App / SDK / MCP Agent ───▶ [ SentioGate Router & /v1/mcp ] ──┬──▶ OpenAI
│ ├──▶ Anthropic
⚡ Cost / quality / latency ├──▶ Google Gemini
🛡️ Automatic fallback ├──▶ DeepSeek
🔒 Encrypted key Vault └──▶ Groq / Mistral / xAI / ...
- 0% Token Markup: SentioGate is Bring-Your-Own-Key (BYOK). You pay your providers' standard rates directly; SentioGate adds no fee per token.
- Routing you control: send
model: "auto"and chooseroute_preferencecost(default),qualityorlatency, or pin a specificprovider/model. - Drop-in OpenAI Compatibility: works with the OpenAI SDK, LangChain and other OpenAI-compatible tools by changing the base URL and API key.
- Automatic fallback: SentioGate automatically falls back to another model if one fails, so a single provider outage or rate limit doesn't have to break your app.
- Your keys stay yours: provider keys are stored encrypted in your Vault, decrypted only in memory to call the provider, and shown back masked.
Connect Claude Code, Claude Desktop, Cursor, Cline, Grok, or Hermes Agent directly to SentioGate's FinOps & Routing Control Plane:
- NPM Package:
@sentiogate/mcp(npx -y @sentiogate/mcp) - Streamable HTTP Endpoint:
https://api.sentiogate.dev/v1/mcp(POST,GET,DELETE— MCP2025-03-26&2024-11-05) - Official MCP Registry:
io.github.SentioGate/sentiogate-mcp - Source Repository: SentioGate/sentiogate-mcp
{
"mcpServers": {
"sentiogate": {
"command": "npx",
"args": ["-y", "@sentiogate/mcp"],
"env": {
"SENTIOGATE_API_KEY": "sg_live_your_gateway_key"
}
}
}
}| MCP Tool | Purpose |
|---|---|
sentiogate_audit_waste |
Audit JSONL/CSV/JSON logs against your enabled Vault models to quantify recoverable spend |
sentiogate_get_finops_ledger |
Inspect live provider spend, key health, cooldowns, and free pool status |
sentiogate_recommend_model |
Rank candidate models by cost, quality, or latency using empirical task & TTFT stats |
sentiogate_guard_check |
Pre-flight tool calls and shell commands for destructive actions and runaway loops |
sentiogate_optimize_payload |
Simulate context compaction & tool-output pruning to measure exact token reduction |
sentiogate_delegate_prompt |
Delegate sub-tasks through model: "auto" or enqueue async community batches |
Explore our open-source developer libraries and resources:
- sentiogate-mcp — Official Model Context Protocol (MCP) Server & FinOps Control Plane (
@sentiogate/mcp). - sentiogate-js — Lightweight, zero-dependency TypeScript & Node.js client SDK (
@sentiogate/sdk). Supports streaming and opt-in retries. - sentiogate-python — Pure standard-library Python SDK (
sentiogate-python) with streaming support. - llm-routing-benchmarks — Small offline synthetic benchmark demonstration; live metrics are on the benchmarks dashboard.
- Developer Hub & Playground — Interactive visualizer showing routing & MCP tools in action.
We standardized on the Nous Hermes Agent harness (replacing Pi) for both Senti Bot (our standalone BYOK personal & multi-channel assistant) and our Autonomous Coding Dashboard:
- 5.2× Faster Agent Routing: 82ms cold TTFT (20ms cached) vs OpenRouter's 850ms in head-to-head Hermes benchmark runs.
- Self-Improving Skills & Native MCP: Combines Hermes Agent's closed learning loop (
agentskills.io), FTS5 cross-session memory, and multi-platform gateway with@sentiogate/mcpFinOps guardrails.
import os
from sentiogate import AutoRouterClient
client = AutoRouterClient(api_key=os.environ["SENTIOGATE_API_KEY"])
response = client.chat(
model="auto", # SentioGate picks the model
messages=[{"role": "user", "content": "Explain quantum computing in 10 words."}],
extra_body={"route_preference": "latency"}, # "cost" (default), "quality" or "latency"
)
print(response["choices"][0]["message"]["content"])import { AutoRouter } from "@sentiogate/sdk";
const router = new AutoRouter({
baseUrl: "https://api.sentiogate.dev",
apiKey: process.env.SENTIOGATE_API_KEY,
});
for await (const chunk of router.streamChat({
model: "auto",
messages: [{ role: "user", content: "Optimize this SQL query..." }],
extraBody: { route_preference: "cost" },
})) {
process.stdout.write(chunk.choices?.[0]?.delta?.content || "");
}import os
from openai import OpenAI
# Just update base_url and api_key
client = OpenAI(
base_url="https://api.sentiogate.dev/v1",
api_key=os.environ["SENTIOGATE_API_KEY"],
)We believe in open metrics. View provider latency and comparative benchmarks:
- 🌐 Live Web Benchmarks: sentiogate.app/benchmarks
- 📈 Benchmark Methodology & Code: SentioGate/llm-routing-benchmarks
- ⚖️ Comparison Guide (SentioGate vs OpenRouter & LiteLLM): sentiogate.app/compare
- Web Application & Dashboard: https://sentiogate.app
- Interactive API & MCP Documentation: https://sentiogate.app/docs
- Senti Bot & Coding Dashboard (Hermes Harness): https://sentiogate.cloud
- US Developer & Latency Hub: https://sentiogate.dev
- UK & EU FinOps & GDPR Hub: https://sentiogate.xyz
- India Regional INR & BYOK Hub: https://sentiogate.com
- Free Chatbot & Gateway Tools: https://sentiogate.app/free-ai-tools
- GitHub Pages Developer Hub: https://sentiogate.github.io
- Developer Inquiries: support@sentiogate.app