One API for every model

Test hundreds of AI models without rewriting your app.

Keep the OpenAI client you already use. Change one base URL, run the same prompt across open and frontier models, and compare the result on price, speed, privacy, and reliability.

Browse live models

No subscription. No card required. Text models cost the provider price plus 5%.

100sof model routes

OneOpenAI compatible client

Zeroprompt or output logs

Keep your clientOne base URL replaces separate integrations for every provider.

Measure, do not guessLive prices, latency, throughput, and availability.

Ship with fallbackHealthy routes can roll over when a provider fails.

Privacy with proofOpen source gateway code, a measured image, and fresh attestation.

Your first useful test

Run a three model bakeoff in 60 seconds.

Do not choose a production model from a leaderboard alone. Send the same prompt from your application to three routes and inspect the answers yourself.

Create an API key and export it as TRUSTEDROUTER_API_KEY.
Install the OpenAI client: pip install openai.
Run the script at right with a prompt from your real workload.

Read the migration guide

Start with aliases, then pin the exact model and provider combination that wins your eval.

Three model bakeoffPython

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["TRUSTEDROUTER_API_KEY"],
    base_url="https://api.trustedrouter.com/v1",
)

prompt = "Write a retry helper with exponential backoff."

for model in (
    "trustedrouter/fast",
    "trustedrouter/cheap",
    "trustedrouter/zdr",
):
    response = client.chat.completions.create(
        model=model,
        messages=[{"role": "user", "content": prompt}],
        max_tokens=256,
    )
    print(f"\n--- {model} ---")
    print(response.choices[0].message.content)

Why change the base URL?

Use competition between models instead of betting on one vendor.

Model choice

Stop integrating providers one at a time.

Use one request format across hundreds of open and frontier model routes. Compare context, capabilities, privacy, and current pricing before you call.

Browse live models → Measured performance

Stop choosing from marketing claims.

Use routed samples to compare latency, throughput, availability, and cost. Then run a small eval on the task your users actually care about.

Open the leaderboard → Production reliability

Stop making one provider your single point of failure.

Use ranked route candidates and provider fallback. Inspect router-core health separately from downstream model availability.

Check live status →

A better model decision

Evaluate on your work, not somebody else's average.

Start with three prompts that represent real success and failure.
Compare trustedrouter/fast, trustedrouter/cheap, and a model picked from the live catalog.
Measure answer quality, total cost, first-token latency, and failures.
Pin the winner, keep a fallback, and expand the eval only when it changes a decision.

Run a focused eval

Privacy with proof

Verify the router. Choose the provider boundary.

TrustedRouter publishes its gateway source, measured workload image, and fresh attestation bound to your nonce. You can verify what runs before sending a prompt.

TrustedRouter never logs prompt or output content. Ordinary synchronous and streaming inference does not retain it. The opt-in Batch API temporarily retains enclave-encrypted artifacts for up to 30 days. Downstream retention, training, jurisdiction, and confidential-compute posture remain provider specific, so each route publishes those facts instead of flattening them into one vague promise.

Verify gateway Provider posture Model routes Source

Run one prompt through three routes. No migration project. Keep your client and compare models on your own task.

No subscription. Delete the key whenever you want.

Questions

Will my existing OpenAI integration work?

In most applications, yes. Keep the OpenAI SDK and request shape, replace the base URL, and use a TrustedRouter model ID. The migration guide documents the compatibility surface and the few differences by provider.

Which model should I try first?

Use trustedrouter/auto for general routing, trustedrouter/fast when latency matters, trustedrouter/cheap when cost matters, or trustedrouter/zdr when zero data retention is required. Then compare the same real prompt across routes before choosing production defaults.

Does TrustedRouter store prompts or outputs?

TrustedRouter never logs prompt or output content. Ordinary synchronous and streaming inference does not retain it. The opt-in Batch API temporarily retains enclave-encrypted artifacts for up to 30 days. Operational metadata includes model, provider, token counts, latency, cost, status, and region. Downstream provider handling remains provider specific and is published on model and provider pages.

What does gateway attestation prove?

A fresh nonce challenge lets you verify that the live prompt gateway is running the published workload image. It does not prove the code is bug free or make every downstream provider confidential.