Sep 2026
Accounts
Sign up with email and password; API keys and usage are now private to your account.
New — Adaptive mode picks the model for you
One key. Every model. Routed for the task.
TrivaNexus is the model gateway for AI agents: an OpenAI-compatible API that sends each request to the model best suited to it, fails over when providers break, and shows you exactly where every token went.
typescript
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://agentic.trivanexus.com/api/v1",
apiKey: process.env.TRIVA_API_KEY,
});
// "adaptive" picks the right model for every request
const res = await client.chat.completions.create({
model: "adaptive",
messages: [{ role: "user", content: "Refactor this function" }],
});19+ models from
OpenAIAnthropicGoogleDeepSeekxAIMistralMetaQwenHuggingFaceKimi
Capabilities
The gateway your agents can lean on
Everything between your agent and the model — routing, reliability, tools and control — in one endpoint.
Adaptive routing
Every request is scored for complexity and sent to the right tier — quick fixes to fast models, architecture work to frontier ones. One conversation stays on one model so caching keeps working.
Automatic failover
If a provider errors, the next one is tried. When nothing can answer you get a clear 502 or 503 — never a fake reply dressed up as a real one.
Built for agents
Streaming, tool calling and a gateway-side file tool loop in a sandboxed workspace, so coding agents like Cline work out of the box.
Organization controls
Approve which models your team can use, set a default, and tune the adaptive tiers and strategy from the admin console.
Usage you can see
Per-key daily budgets, token and cost metering, and a live view of which models adaptive mode actually routed to.
How it works
From sign-up to routed requests in four steps
No SDK to learn and nothing to self-host.
- 01
Create an account and a key
Sign up, open the dashboard and create an API key. Keys are yours alone and can be revoked any time.
Create account → - 02
Point your client at TrivaNexus
Change the base URL of any OpenAI-compatible SDK or agent. No new libraries, no new request format.
Read the docs → - 03
Let adaptive pick the model
Request model "adaptive" and each turn is routed to a model that can serve it — escalating a tier if a call fails.
Browse models → - 04
Watch and tune
Track usage per key, see routed models per tier, and adjust budgets and allow-lists as your team grows.
Open dashboard →
Integrations
Works with the tools you already use
Anything that speaks the OpenAI chat completions API can use TrivaNexus — just change the base URL.
Cline
VS Code coding agent
Continue
IDE assistant
OpenAI Node SDK
baseURL
OpenAI Python SDK
base_url
LangChain
ChatOpenAI
cURL / HTTP
/v1/chat/completions
OpenRouter
hundreds of upstream models
Local models
llama.cpp · Ollama · vLLM
Changelog
Recently shipped
Sep 2026
Adaptive mode
One model id that routes each request to the right tier and stays consistent across a conversation.
Sep 2026
Gemini 3.8 Flash, Gemini 3.1 Pro, Grok 4.7
Current-generation models, with old ids kept as aliases.
Sep 2026
Honest errors
Failed or unconfigured providers return 502/503 instead of placeholder text.
Security
Safe by default
Guardrails for your accounts, keys and agents are built in, not bolted on.
Hashed passwords
Account passwords are stored as salted scrypt hashes.
Hardened sessions
httpOnly, SameSite cookies; only a hash of each session token is stored.
Private API keys
Keys are generated from a secure random source and scoped to their owner.
Budgets per key
Daily token limits stop a runaway agent from draining your balance.
Model allow-lists
Admins decide which models can be used across the organization.
Sandboxed agent tools
Gateway file tools are jailed to a workspace; path traversal is blocked.
Give your agents the right model, every time.
Free to start. Bring your own client, keep your code.