TrustedRouter Review (2026): Pricing, Features & Honest Verdict

TrustedRouter is an AI model gateway: one OpenAI-compatible API for 625 models across 91 providers, routed through an attested enclave that keeps no prompt or output logs. We found the privacy engineering unusually checkable and the markup identical to OpenRouter’s 5.5%. Best for: teams routing sensitive prompts. Price: provider cost plus 5.5% (free plan: no). Rating: 7.5/10.

What is TrustedRouter?

TrustedRouter is a model gateway. You change one base URL in the OpenAI SDK and your application can then call 625 models from 91 providers through a single key, with failover, regional routing and price controls handled on the way. That part is not new. The part that is new is what the gateway does with your prompt while it is passing through.

A gateway sits at the exact point in your stack where every prompt and every completion has to travel. Most of them log that traffic, because logs are how you build analytics dashboards and how you debug. TrustedRouter runs the prompt path inside a Google Confidential Space workload, terminates TLS inside the measured enclave, and publishes the source commit, the container image reference and the image digest so you can compare what is running against the open-source repository. It launched on Product Hunt on 1 September 2026 under the tagline “Every model with a unified interface. Privacy with proof.”

The company is Lore Hex Corp, a Delaware C corporation at 1111 Brickell Ave in Miami, founded by Joseph Perla. Its about page publishes the EIN, the DUNS number and a business phone, which is not something we see on a launch-day product more than a couple of times a year. Stefan Heinrich and Sam Lessin are credited alongside Perla as makers. The positioning is stated outright rather than implied: the site calls itself an OpenRouter alternative and the docs carry a page titled “vs OpenRouter.”

What are TrustedRouter’s key features?

An attested prompt path you can check yourself

The trust page is the best thing on the site. It names the current GCP workload by source commit, image reference and sha256 digest, then gives you five steps to verify it: fetch the live attestation from the API over normal TLS, check the JWT issuer is Google’s confidential computing service, check the audience, compare the attested digest against the one published on the page, and confirm the TLS certificate fingerprint is bound into the attestation nonce. That last step is the one that matters, because it means the connection you are on is the connection that was attested rather than some other connection that happened to pass a check.

There is a real gap, and TrustedRouter discloses it rather than hiding it. Only the GCP plane has published measurements. The AWS Nitro plane lists PCR0 as “not-configured” and the Azure Confidential Containers plane has no hostdata value. The page tells you plainly not to treat the absence of a measurement as a measurement of zero. We respect that a lot more than a green tick, but it does mean that “attested” today describes one of three advertised planes, and the headline does not carry that qualifier.

625 models behind one base URL

The catalog is public and it is worth opening even if you never sign up. Every row carries input price, cached input price, output price, context window, provider count and privacy labels, and the filters include open weights, US providers, EU-focused, ZDR available and E2EE available. You can sort the entire catalog by lowest cached-input price, which is the number that actually decides the bill on a long-context agent and which almost no vendor lets you sort by.

Prices are visibly derived rather than negotiated. Claude Sonnet 5 lists at $2.11 in and $10.55 out per million tokens, against Anthropic’s $2 and $10. Claude Opus 5 lists at $5.275 and $26.375. Gemini 3.5 Flash lists at $1.5825 input. Every one of those is the provider’s own price multiplied by 1.055, which makes the arithmetic checkable from the outside in about ten seconds.

Routing aliases that encode a privacy policy

Instead of a settings toggle, the privacy posture is a model name. Send a request to trustedrouter/zdr and you get a route with a tracked zero-data-retention guarantee. Send it to trustedrouter/e2e and you require both provider-side confidential compute and end-to-end encryption. There is trustedrouter/eu for European routing and trustedrouter/auto for a healthy route at the selected provider’s price.

The behaviour under failure is the right one. The docs describe these as hard minimums: if no endpoint matching your model and provider filters carries the required guarantee, the request fails before inference rather than silently downgrading to a route that does not meet the bar. Silent downgrade is the standard failure mode in this category and it is the one that gets a compliance team fired.

BYOK, and a marketplace where you can sell your own endpoint

You can bring your own provider keys, which is the sensible path if you already hold committed spend or enterprise rate limits somewhere and want the attested path and the failover without giving up your discount. There is also a user-provided models feature that runs the other way: you list your own endpoint, a machine, an agent or a person typing, at a price you set, and keep 70% of what it earns in credits.

Hold that 70% next to the 5.5% and you have the clearest picture of how this business works. Buying inference costs you 5.5% over the provider price. Selling it costs you 30%. Neither number is hidden, but only one of them is on the pricing page.

How much does TrustedRouter cost?

TrustedRouter has no subscription and no seat fee. Text and embedding routes bill at the provider’s token price plus 5.5%, with a floor of $0.01 per million tokens. Video generation bills at the provider’s exact per-job quote plus 20%. You fund it with prepaid credits by card, ACH, PayPal or stablecoin, or you bring your own key, and everything is measured in integer microdollars even when the interface rounds. Teams already spending more than $100 a month on LLMs can request migration credits, subject to approval.

The comparison the site invites is with OpenRouter, so we ran it. OpenRouter charges 5.5% with a $0.80 minimum on credit purchases through Stripe, 5% on crypto, and 5% of the equivalent cost when you use your own key. TrustedRouter charges 5.5% at request time instead of at top-up time, which nets out the same on card payments. On stablecoin it does not: OpenRouter takes 5% there and TrustedRouter still takes 5.5%, so paying in crypto is 10% more expensive per dollar routed here than at the incumbent.

Two more differences move real money. The $0.01 per million floor means free models are not free on this gateway. OpenRouter publishes a set of models at $0 with rate limits, and that free tier is how a lot of side projects and evaluations start. Here the cheapest possible route costs a cent per million tokens, which is trivial in absolute terms and still removes the zero. And the 20% video markup is 3.6 times the text markup, disclosed in the body copy but absent from the headline, so a team that reads only the top of the page will price a video workload wrong.

The gap we could not close is BYOK. The pricing page lists it as one of three ways to pay and describes what you keep, and it never says what it costs. OpenRouter publishes 5% for the same arrangement. A missing number on the page where the numbers live is the one thing on this site that does not match the standard the rest of it sets.

Who is TrustedRouter best for?

Use TrustedRouter if your prompts carry anything you would have to disclose in a breach notification. Legal documents, health records, customer support transcripts, internal code. The whole product is built for the buyer who needs to answer “where did the prompt go” with evidence rather than with a vendor promise, and the attestation flow gives that buyer something to hand a security reviewer. The published company record helps too: a procurement team can look up the EIN and the DUNS without emailing anyone.

Use it as well if you are already on OpenRouter and want the same economics with a verifiable prompt path. The migration is a base URL change, the markup matches on card payments, and the catalog is larger.

Skip TrustedRouter if you are a hobbyist or evaluating models on a weekend. The lack of a free tier and the $0.01 floor mean OpenRouter’s free models remain the cheaper on-ramp. Skip it if your workload is video-heavy, where the 20% markup starts to hurt at volume and going direct is worth the integration work. And skip it if you need a contractual SLA today, because the terms state that no service-level agreement applies unless you sign a separate agreement, and the liability cap is the greater of $100 or twelve months of fees.

Best TrustedRouter alternatives

OpenRouter is the incumbent and the honest first choice for most teams. Same 5.5% headline on card top-ups, 5% on crypto, 5% BYOK, a real catalog of free models with rate limits of 50 requests a day or 1,000 once you hold $10 in credits. What you do not get is an attested prompt path or a published image digest, so the privacy posture is a policy rather than evidence.

LiteLLM is the answer if you would rather own the gateway than trust one. It is open source and free to run yourself, which removes the markup entirely and moves the privacy question inside your own perimeter. The trade is that you now operate a piece of infrastructure, handle provider failover yourself, and pay in engineering time instead of basis points.

Portkey comes at the same problem from the observability side, with a free forever tier covering 10,000 recorded logs a month and a paid plan at $49 a month for 100,000. It is the better pick if what you actually want is gateway analytics, prompt versioning and a playground, and the weaker pick if the reason you want a gateway is that you do not want your prompts recorded at all.

Final verdict: is TrustedRouter worth it?

Yes, for the buyer it was built for. TrustedRouter charges the same 5.5% as the market leader and hands you something the market leader does not offer: a live attestation, a published image digest, an open gateway repository and a documented way to check that the three agree. The routing aliases fail closed instead of downgrading, the model catalog is the most inspectable one in the category, and the company publishes its EIN, DUNS and registered address on a page anyone can read. That combination is rare enough that we would move a sensitive workload here before we would move a cheap one.

The reasons to wait are specific rather than structural. Two of the three advertised trust planes have no published measurements yet, which the site admits and which still leaves “attested” doing more work in the marketing than in the AWS and Azure release records. The BYOK price is missing from the pricing page. Video costs 20% rather than 5.5%. And there is no free tier at all, so evaluating this gateway costs money on the first request in a category where the incumbent lets you start at zero. Fix the BYOK number and publish the other two attestations and this is an easy 8. Today it is a 7.5, and the best-documented launch we have reviewed this month.