Miami’s TrustedRouter Raises $6M to Keep AI Prompts Private

Brickell-based TrustedRouter raised a $6 million seed to route AI requests across hundreds of models while keeping prompts private, six weeks after launch.

Every AI prompt travels through someone’s servers. Usually, that someone can read it.

Brickell-based TrustedRouter has just raised $6 million to build the opposite. Semafor reported that Sierra Ventures led the seed round, six weeks after the company launched publicly.

What Does a Router Actually Do?

A router sits between a company and the AI models it uses. It sends each request to the model that best fits the cost, speed and privacy the job needs.

That lets developers switch models without rebuilding their applications. Founder Joseph Perla says no one should be locked in to a single provider.

TrustedRouter gives developers access to hundreds of models through one connection, and its routing code is open source.

Perla targets businesses whose AI spending is growing, Semafor says. They want to keep their data private.

The twist is privacy. For some open-weight models, TrustedRouter offers end-to-end encryption and verifiable confidential computing. Neither the company nor the data center can read the prompts or the answers.

Semafor notes that closed models from OpenAI and Anthropic do not offer the same guarantees. Neither does OpenRouter, the category’s largest player.

Why Is Privacy the Pitch?

Perla is a former Facebook engineer and early Lyft hire who has founded and sold several startups, Semafor says.

OpenRouter is the competitor everyone measures against. It started with a $12.5 million seed round in February 2025. It has now agreed to sell to Stripe for a reported $7.5 billion.

Perla is betting that routing alone is not enough. Privacy is his edge.

“Security is just a pure baseline. It just has to be secure,” he told Semafor.

MAIN covered the first seed round in A Brickell startup just raised $1.25M for AI privacy. Hundreds of customers had joined by then, Perla said, many from legal, finance and healthcare.

Those are fields where client data carries real regulatory weight.

Legal work is a good example. MAIN reported on Miami Law faculty writing AI’s rulebook, a local effort to set rules for AI in courts and firms.

Healthcare is the sharper one. MAIN covered OpenEvidence going inside Northwestern Medicine’s EHR, where an AI tool now works next to patient records.

Any tool in that position raises the question TrustedRouter is selling against. Who else can see what the model sees?

Encryption has a second effect, Perla argues. If enough companies encrypt their AI chats, frontier labs get less user data to train on.

Perla has bigger plans than privacy. He wants to launch a marketplace called the New York Token Exchange.

In it, models and hosting providers would compete to serve each request. Perla pictures AI tokens traded like oil or electricity, with futures and price indices.

He has also floated “model IPOs” for new open-weight releases. Chinese models from DeepSeek, Zhipu and Moonshot are among the most popular on his platform, Semafor reports.

Perla says he wants a more vibrant American open-source ecosystem. Semafor notes new open-weight releases from Reflection AI and Mistral. Meta calls its newest the most capable US open-weight model to date.

A Brickell Neighborhood Taking Shape

TrustedRouter now shares its neighborhood with big AI money. Hut 8 is headquartered at 777 Brickell Avenue. MAIN covered Miami’s Hut 8 landing a $1.07B AI data center loan.

Citadel is building its tower on Brickell Bay Drive. The neighborhood is collecting the people who finance AI and, now, a startup that sits between users and the models.

Router software decides which model sees which data. That is why the layer matters so much for privacy.

A seed this size, six weeks after launch, is a vote that privacy can set a router apart. A Brickell startup now has the funding to prove it with enterprise customers.

Sources: