There is a slightly uncomfortable problem with the rise of AI routers.

The whole point is to stop trusting one AI company with everything.

So you insert another company directly between your application and every AI model you use.

Now that company can potentially see everything instead.

TrustedRouter thinks that is the wrong trade.

The Miami AI infrastructure startup offers developers one OpenAI-compatible API that can route requests across hundreds of models and providers. But its more interesting pitch is not convenience.

It is this:

You should not have to trust the router either.

TrustedRouter says prompt traffic passes through an attested confidential-computing gateway that does not log prompt or output content, with source code and runtime measurements available for customers to inspect.

And investors have started paying attention.

On August 31, 2026, TrustedRouter announced a $1.25 million seed round. Axios reported that the company had recently processed more than 1 billion tokens in a single day, while TrustedRouter says it is bringing on hundreds of customers.

So is this another API middleman riding the AI infrastructure boom?

Or is TrustedRouter building something that could become genuinely important?

What’s Going On With TrustedRouter?

TrustedRouter was founded in 2026 by Joseph Perla and is operated by Lore Hex Corp, a Delaware corporation headquartered in Miami. The product entered public beta in May.

The basic proposition is straightforward.

Instead of integrating OpenAI, Anthropic, Google, DeepSeek, Mistral and every other provider individually, developers can point an OpenAI-compatible client toward TrustedRouter.

From there, they can select individual models or ask TrustedRouter to handle parts of the routing process.

Its current website advertises:

Metric	Current public figure
AI models	600+
Providers	90+
Clouds	AWS, Azure, Google Cloud
Seed funding	$1.25M
Public beta	Since May 2026
Peak milestone	1B+ tokens in one day
Standard text/embedding fee	Provider cost + 5.5%
Monthly subscription	None

The 600+ models and 90+ providers figures are current company claims. Axios independently reported more than 600 models and more than 81 providers on August 31, indicating the catalog was already substantial before the newer count appeared.

No public valuation or audited revenue figure was found during our research.

That matters.

This is a very early infrastructure company, not another $10 billion AI unicorn.

Who Founded TrustedRouter?

TrustedRouter is led by Joseph Perla, whose background is considerably more substantial than the company’s small seed round might suggest.

Perla studied computer science and machine learning at Princeton, founded Turntable.fm, worked on Facebook’s iPhone News Feed, co-founded Terminal.com, which was later acquired by Udacity, and worked at Lyft during its growth years. His own biography also describes work on passenger and driver growth at Lyft.

That background matters because TrustedRouter is not primarily a consumer AI application.

It is infrastructure.

The company has to deal with provider integrations, authentication, distributed systems, billing, failover, security boundaries, latency and confidential computing.

Perla has also described TrustedRouter as an unusually AI-native company, saying he built much of the initial system while working with AI coding tools rather than assembling a conventional large startup team. That is a founder claim, but an interesting one given the scope of the public repositories.

The Funding

TrustedRouter announced a $1.25 million seed round on August 31, 2026.

The most prominently disclosed investors include:

• Sam Lessin of Slow Ventures
• Bill Tai
• Linda Avey, co-founder of 23andMe
• George Xing
• Peter Livingston of Unpopular Ventures
• Katelyn Donnelly of Avalanche VC
• Gert Lanckriet
• Holmes Wilson
• Tory Reiss
• Daniel Imberman
• Michael Staton
• Jason Fang
• Capitoria Ventures
• other technical and angel investors

The company also identified Anthropic growth engineer Alexey Komissarouk and former OpenAI engineer Henri Roussez among participants.

CB Insights independently records the $1.25 million financing, although its simplified investor record lists only the four headline investors.

Valuation: not publicly disclosed.

That is worth emphasizing because there is no need to manufacture a giant valuation story here.

TrustedRouter is interesting because of the infrastructure thesis, not because somebody attached ten zeroes to it.

What Does TrustedRouter Actually Do?

Imagine your application currently calls OpenAI directly.

Normally, your architecture might look like:

Your app → OpenAI

Then you decide Gemini is better for one job.

Claude is better for another.

A cheaper open model can handle classification.

Another provider is faster in Europe.

And your primary provider occasionally goes down.

Suddenly you are maintaining multiple integrations, billing systems, credentials and fallback rules.

A router inserts one abstraction layer:

Your app → router → whichever model/provider is appropriate

TrustedRouter uses an OpenAI-compatible interface, meaning many applications can migrate primarily by changing the API base URL rather than rewriting their entire model integration.

Developers can select individual models, but TrustedRouter also exposes routing aliases.

For example, its documentation currently includes routing options designed around automatic provider fallback, zero-data-retention requirements, confidential-compute requirements and EU-focused routing.

That means the router is not merely answering:

Which model?

It can increasingly answer:

Which model, from which provider, in which region, under which privacy constraints?

That is a much more interesting infrastructure position.

Why Privacy Is the Real Product

An AI router occupies an uncomfortable place in the stack.

Every prompt can pass through it.

Every response can pass back through it.

For an AI coding agent, that might mean proprietary source code.

For a law firm, privileged documents.

For a healthcare product, sensitive patient information.

For an enterprise agent, internal strategy, customer records or financial information.

TrustedRouter’s answer is confidential computing.

Its current GCP prompt path runs in Google Confidential Space, and its trust page publishes the source commit, container image, image digest and instructions for checking live attestation. Public TLS is designed to terminate inside the measured workload rather than in a normal external control plane.

TrustedRouter says its ordinary inference path stores metadata such as model, provider, token counts, cost, region and status, but not prompt or response bodies, according to its security documentation.

That is materially different from simply publishing a privacy policy saying:

“We promise not to look.”

Attestation is designed to let someone verify that the running workload corresponds to a known software image.

And that is probably TrustedRouter’s strongest idea.

But “Private” Has an Important Boundary

This is where the story gets more nuanced.

TrustedRouter can provide evidence about its gateway.

That does not magically make every downstream AI provider confidential.

TrustedRouter’s own security documentation acknowledges this clearly: upstream provider retention and training policies remain provider-specific.

That means:

User → TrustedRouter gateway

can have one privacy guarantee.

But:

TrustedRouter → model provider

can have another.

For workloads needing stronger guarantees, TrustedRouter supports routing constraints intended to require providers offering confidential-compute and end-to-end encrypted paths.

Its comparison with Tinfoil is surprisingly candid on this point: Tinfoil verifies deeper into the actual model-serving stack, while TrustedRouter’s advantage is breadth across many more providers and models.

That distinction is critical.

TrustedRouter reduces one trust problem.

It does not erase the entire AI supply chain.

There Is Another Caveat: “Open Source”

TrustedRouter repeatedly describes itself as open source.

That needs an asterisk.

The public repositories absolutely are valuable: developers can inspect substantial portions of the infrastructure and compare running software measurements with published source.

However, the current gateway and control-plane repositories are licensed under Business Source License 1.1.

Under the published terms in the TrustedRouter gateway repository, the source can be inspected, built and used for non-production purposes, while production use requires a commercial license. Each version is scheduled to convert to Apache 2.0 four years after publication. Code released before July 3, 2026 remains Apache-licensed.

Interestingly, TrustedRouter’s own LiteLLM comparison now explicitly calls its gateway “source-available.”

So the fair description is:

TrustedRouter publishes inspectable source, but its current core gateway is not unrestricted open source in the conventional MIT/Apache sense.

That does not undermine the attestation mechanism.

But the terminology should be precise.

The Real Bet: Models Become Commodities

Privacy alone is not the entire TrustedRouter thesis.

The bigger strategic bet is that no single AI model stays best at everything.

And the market is increasingly moving in that direction.

OpenRouter offers hundreds of models and more than 80 providers. Vercel AI Gateway now advertises hundreds of models behind one API with automatic failover. LiteLLM supports more than 100 provider integrations and gives companies a gateway they can self-host.

Even Runway launched a router for generative media in July, explicitly arguing that the explosion of models makes it increasingly difficult for developers to decide which model to use for each job.

The trend is hard to miss.

The model layer keeps fragmenting.

The routing layer becomes more valuable because of it.

AI inference orchestration is becoming its own battleground.

TrustedRouter Is Also Experimenting With Multi-Model Intelligence

TrustedRouter is not limiting itself to switching between providers.

Its Synth system sends a problem to multiple models and then uses judge and synthesis models to combine their work.

In TrustedRouter’s published DRACO evaluations, its best reported multi-model setup scored 73.4, compared with a strongest solo-model result of 63.0 within the company’s test. TrustedRouter has published the evaluation workflow and artifacts for reproduction.

Treat those numbers as company-run benchmark results, not an independent declaration that TrustedRouter has invented a universally superior model.

But the idea is important.

A router could eventually do more than choose one AI.

It could assemble multiple AIs into one system.

That turns routing from plumbing into orchestration.

Does Anyone Actually Use It?

This is still one of TrustedRouter’s biggest unanswered questions.

The company says it is onboarding “hundreds and hundreds” of customers across sectors including legal, finance and healthcare, but it does not publish a comprehensive customer list.

Its strongest public case study is legal AI company Robot, Robot & Human.

According to the companies’ published case study, RRH processed 170,974 documents across 34 matters, totaling more than 616,000 pages, while routing billions of tokens through TrustedRouter. The deployment reportedly moved from first call to production-scale batch processing in roughly three weeks.

Those are company-supplied customer metrics rather than independently audited figures.

Still, they demonstrate something more useful than a vague logo wall: a concrete production workload.

Axios also reported the more significant platform-level milestone that TrustedRouter had recently crossed one billion tokens in a single day.

For a product that only entered public beta in May, that is meaningful early usage.

How Much Does TrustedRouter Cost?

TrustedRouter does not require a monthly subscription.

For normal prepaid text and embedding requests, it currently charges the upstream provider price plus 5.5%, subject to a small minimum.

Video generation carries a 20% markup.

It also supports bring-your-own-key arrangements for teams with existing provider contracts or rate limits, according to TrustedRouter pricing.

That puts TrustedRouter’s standard markup in roughly the same territory as OpenRouter’s published 5.5% fee on standard pay-as-you-go credit purchases.

Vercel is particularly aggressive here: its AI Gateway currently advertises provider list pricing with no token markup, although certain capabilities can be tied to paid Vercel plans or separately metered functionality.

So TrustedRouter cannot win purely by saying:

“Use us because routing is cheaper.”

Its privacy architecture needs to matter.

TrustedRouter vs. OpenRouter, Vercel, LiteLLM and Tinfoil

The competitive landscape is already serious.

OpenRouter

Probably the most obvious comparison.

OpenRouter already offers 500+ models and 80+ providers, extensive routing controls, provider policy filters and substantial developer adoption.

TrustedRouter’s argument is that privacy verification should be stronger.

Vercel AI Gateway

Vercel AI Gateway offers hundreds of models, OpenAI- and Anthropic-compatible migrations, automatic failover, BYOK and zero-markup provider pricing. It also exposes controls around zero-data-retention providers and provider allowlists.

That is formidable competition because Vercel already owns a huge developer distribution channel.

LiteLLM

LiteLLM takes a different approach.

Its MIT-licensed gateway can be self-hosted, giving enterprises direct infrastructure control, with optional enterprise governance and support.

For organizations that already prefer operating their own infrastructure, that may be more attractive than adding another hosted service.

Tinfoil

Tinfoil is arguably TrustedRouter’s most interesting privacy comparison.

Tinfoil runs supported models inside confidential hardware and performs verification down through its enclave chain. That gives it a deeper technical privacy boundary, but around a narrower model universe.

TrustedRouter is essentially betting that developers want both breadth and verifiability.

That is a compelling position.

It is not an uncontested one.

Readers comparing this market can also browse more AI tools and infrastructure reviews on AI Tribune.

Enterprise Buyers Should Notice What Is Missing

TrustedRouter deserves credit for publishing a procurement page that does something many young startups avoid:

It plainly lists what it does not have yet.

As of September 10, 2026:

• its DPA is available but requires signing;
• its subprocessor information is published;
• SOC 2 has not yet been obtained;
• HIPAA readiness material exists, but appropriate agreements such as a signed BAA are still necessary for relevant deployments.

Those details are published in TrustedRouter’s legal and procurement documentation.

That transparency is good.

But a security architecture and a compliance certification are not interchangeable.

A regulated enterprise should still perform normal legal and security diligence.

Is TrustedRouter Actually Legit?

Yes.

TrustedRouter is a real operating company with an identifiable legal entity, founder, public repositories, functioning APIs, public trust infrastructure, outside financing and documented production usage, according to its company information.

But “legit” should not be confused with “proven.”

TrustedRouter is only months into public beta.

No public ARR figure was found.

No valuation has been disclosed.

Independent customer commentary remains limited.

SOC 2 has not yet been obtained.

And the company is competing directly with much larger platforms.

Pros and Cons

Pros	Cons
600+ models behind one API	Very young company
Strong founder background	No disclosed ARR or valuation
Unusually transparent attestation architecture	Limited independent customer corpus
No prompt/output logging at TrustedRouter gateway	Downstream provider privacy still varies
Published source and live trust evidence	Current core is BUSL source-available, despite broader “open source” wording
Automatic routing and provider failover	Vercel, OpenRouter, LiteLLM and others are formidable competitors
BYOK and no required subscription	No SOC 2 report yet
Concrete legal-AI case study	Long-term reliability at much larger scale remains unproven

WHO SHOULD TRY IT?

TrustedRouter is most interesting for developers and companies already using multiple AI models.

Especially:

• AI products using OpenAI-compatible APIs
• coding agents
• document-processing systems
• companies worried about adding another opaque gateway
• teams wanting provider failover
• organizations experimenting with multiple models
• companies needing geographic or privacy-based routing
• developers evaluating an OpenRouter alternative

For a hobby project that makes a handful of API calls every month, much of this infrastructure may be unnecessary.

For a company pushing proprietary information through several AI providers every day, the proposition becomes much more interesting.

Hype Meter: 8.1/10

TrustedRouter has a strong technical thesis, credible founder, real early usage and unusually inspectable privacy architecture, but its enterprise maturity, independent customer evidence and long-term competitive moat still need considerably more proof.

FAQ

What is TrustedRouter?

TrustedRouter is an AI gateway that provides one OpenAI-compatible API for accessing hundreds of models and providers while adding routing, failover and privacy controls.

How many models does TrustedRouter support?

TrustedRouter currently advertises 600+ models and 90+ providers; Axios reported 600+ models and 81+ providers on August 31, 2026.

How much funding has TrustedRouter raised?

TrustedRouter announced a $1.25 million seed round on August 31, 2026.

Who founded TrustedRouter?

TrustedRouter was founded by Joseph Perla, a former Facebook and Lyft engineer and entrepreneur behind projects including Turntable.fm and Terminal.com.

Does TrustedRouter store prompts?

TrustedRouter says its standard synchronous gateway does not store prompt or output bodies and publishes attestation information intended to make that behavior independently verifiable. Some optional functionality, such as batch processing, has different storage boundaries that customers should review separately.

Is TrustedRouter completely private?

Not automatically. TrustedRouter can attest its own gateway, but privacy protections at the downstream model provider remain provider-specific unless stronger route requirements are selected. TrustedRouter explains those boundaries in its security documentation.

Is TrustedRouter open source?

Parts of its ecosystem use permissive licenses, but its current gateway and control plane are published under BUSL 1.1, making “source-available” the more precise description for current releases. Production use of those versions requires a commercial license until their scheduled Apache 2.0 conversion.

How much does TrustedRouter cost?

Text and embedding requests currently cost the underlying provider price plus 5.5%. Video generation is marked up 20%. There is no mandatory monthly subscription.

What are the main TrustedRouter alternatives?

The most relevant alternatives include OpenRouter, Vercel AI Gateway, LiteLLM and Tinfoil, depending on whether the priority is model breadth, self-hosting, price, developer ecosystem or confidential inference.

Final Take

TrustedRouter may be early, but the problem it is solving is getting bigger almost every week.

AI used to be simple.

Pick OpenAI.

Call the API.

Done.

That world is disappearing.

Claude may win one task.

Gemini another.

A Chinese open model may suddenly deliver 80% of the performance for 5% of the price.

A specialist inference provider may be dramatically faster.

A provider may go down.

A security team may ban another.

A European customer may require regional processing.

Once AI applications become genuinely multi-model, something has to sit between the application and that chaos.

That something is the router.

TrustedRouter’s most interesting insight is that the router itself becomes a security problem the moment it becomes important.

So it is trying to make routing inspectable before routing becomes invisible infrastructure.

There are plenty of reasons it could lose.

OpenRouter already has scale.

Vercel owns developer distribution.

LiteLLM lets enterprises own the gateway themselves.

Tinfoil goes deeper on confidential inference.

And the frontier AI providers themselves have every incentive to make switching away from them unnecessary.

But TrustedRouter does not need every AI application on Earth.

It needs one broader prediction to come true:

There will not be one model that wins everything.

If that happens, model selection becomes infrastructure.

And if increasingly sensitive information flows through that infrastructure, trust cannot just be another paragraph in a privacy policy.

That is where TrustedRouter gets genuinely interesting.

The company is not really betting that developers want another AI API.

It is betting that eventually they will want one layer that lets them use every AI without blindly trusting the layer in the middle.

That is a much bigger idea.