Find a model

Browse the full catalog →

One key, one model, one provider

TrustedRouter is a public hosted API at api.trustedrouter.com, independent of OpenRouter. It provides an OpenAI-compatible interface to named upstream providers through an attested gateway.

Create an inference key in the console using the Create key button above, fund credits, and set TRUSTEDROUTER_API_KEY in your environment. These examples call DeepSeek V4 Pro on DeepSeek. provider.only pins the provider and provider.allow_fallbacks: false disables fallback to another candidate. If that route cannot serve the request, it fails. No automatic model selection or provider fallback is enabled by these examples.

curl

curl https://api.trustedrouter.com/v1/chat/completions \
  -H "Authorization: Bearer $TRUSTEDROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v4-pro",
    "messages": [{"role": "user", "content": "Hello"}],
    "provider": {"only": ["deepseek"], "allow_fallbacks": false}
  }'

Python with the OpenAI SDK

Install the SDK with pip install openai. Pass TrustedRouter's provider controls in extra_body.

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["TRUSTEDROUTER_API_KEY"],
    base_url="https://api.trustedrouter.com/v1",
)
response = client.chat.completions.create(
    model="deepseek/deepseek-v4-pro",
    messages=[{"role": "user", "content": "Hello"}],
    extra_body={
        "provider": {"only": ["deepseek"], "allow_fallbacks": False}
    },
)
print(response.choices[0].message.content)

The OpenAI SDK uses normal HTTPS and does not verify gateway attestation automatically. See gateway verification for the checks and their limits. Pinning DeepSeek does not itself impose ZDR or confidential compute; use the privacy filters with an eligible route when required.

What you can call through the same key

  • Chat: POST /v1/chat/completions.
  • Embeddings: POST /v1/embeddings, including openai/text-embedding-3-small.
  • Transcription: POST /v1/audio/transcriptions, including openai/whisper-large-v3-turbo on Tinfoil.
  • Batch: asynchronous requests with a separate encrypted retention boundary.
  • Video: asynchronous generation through POST /v1/videos.

Supported inputs, route controls, and retention differ by endpoint. Check the linked documentation before reusing chat settings.

What TrustedRouter is not

  • There is no reranking endpoint today.
  • TrustedRouter does not offer fine-tuning.
  • It does not run inference GPUs itself. Routes go to named upstream providers.
  • Gateway attestation does not make every upstream provider confidential. Who can see your data explains each party and its retention boundary.

Even for a single-model app, TrustedRouter provides one key and one bill across 600+ models and 100+ providers, per-key spend controls, opt-in provider failover from this pinned setup, and a verifiable gateway. Switching to a compatible model or provider is a one-string change to the model ID or provider selection, with no new provider account; keep any provider pin and privacy filters compatible with the route you select.

A gateway adds a network hop and a service dependency. Pinning removes provider selection and fallback, but it does not remove that hop. TrustedRouter is your contact for its hosted API, routing, and billing; upstream availability and processing remain dependencies. Use the support path for incidents and the business terms to establish responsibilities in writing.

When you outgrow one model

To enable provider failover from this setup, set provider.allow_fallbacks: true and expand or remove provider.only. Use provider.order to prefer providers. To enable automatic model selection, choose trustedrouter/auto and review its eligible pool. Privacy and provider filters still apply.

Fallbacks default to enabled when you omit the setting, so keep the explicit false for a single-provider app. The routing guide explains these choices.

provider.preferred_max_latency and provider.preferred_min_throughput are preferences, not guarantees or timeouts. Pinning gives you control over the provider, not a latency guarantee. Consult performance routing, status, and published benchmark reports for measured behavior.

Capacity and fee

The New York Token Exchange sells 2027 reserved tokens by card. For larger capacity arrangements, contact the team. A reserved-token purchase is not a claim that TrustedRouter operates dedicated GPUs.

The standard platform fee is 5.5% on provider prices, the same rate as OpenRouter's standard gateway fee. It pays for the hosted gateway, unified API and billing, and available routing controls. With a single direct provider, you would not pay this platform fee and would not get these TrustedRouter features. See pricing for minimum charges and optional service pricing.

Business terms and evidence

SOC 2 is in its observation window. No SOC 2 certification is claimed. SLA, MSA and DPA are available to business customers on request through contact. Review the legal packet, DPA, and SOC 2 readiness page; confirm scope and uptime commitments in the written agreement.

The privacy policy sets out retention terms. The trust page supplies verifiable gateway release evidence. The status page reports service health; it is not an uptime SLA.

The homepage links coverage in Axios and Semafor, and names Robot, Robot & Human as a customer. Press coverage and a customer story provide context, not a security certification or performance guarantee.

Workspace access

Sign in

Choose a sign in method to access your TrustedRouter workspace.

New to TrustedRouter?

By continuing you agree to the terms of service and privacy policy.