TrustedRouter raises a $6 million seed to compete with OpenRouter

TrustedRouter has closed a $6 million seed round led by Sierra Ventures. The company told Semafor of the close on 16 October 2026. TrustedRouter is a model-routing service that lets developers and companies send requests to different AI models through one interface. It launched publicly about six weeks earlier. No valuation, no full investor list beyond the lead, and no revenue figure were disclosed. The company positions itself against OpenRouter. OpenRouter raised a $12.5 million seed in February 2025 and later agreed to sell to Stripe in a deal valued at about $7.5 billion. TrustedRouter’s chief executive, Joseph Perla, says the difference is security. The service offers end-to-end encryption and verifiable confidential computing for some open-weight models. In those routes neither TrustedRouter nor the data centre running the model can read a customer’s prompts or outputs. Perla has said OpenRouter does not offer the same protections. Closed models from OpenAI and Anthropic do not currently offer the same guarantees either. Perla is a former Facebook engineer and an early hire at Lyft. He has founded and sold other startups. TrustedRouter had previously raised $1.25 million in a seed round reported in late August 2026. That earlier round included Sam Lessin of Slow Ventures, Bill Tai, Linda Avey and George Xing. The company has said its code is open source and that it gives access to hundreds of models from many providers. Chinese models from DeepSeek, Zhipu and Moonshot have been among the most popular on the platform. A compliance officer at a bank in Riyadh faces a concrete limit when routing AI traffic. Under Saudi data rules, raw customer prompts that leave the Kingdom for a third-party router create a residency and audit question. TrustedRouter’s confidential-computing routes are designed to keep the prompt unreadable by the router and by the data centre. Those routes exist only for some open-weight models. A bank that needs the guarantee must select a supported route and confirm the attestation. The company has not said how many of its listed models offer that path, or what share of its traffic uses it. Without those numbers the product remains a description rather than a control a regulator can examine. Model routing itself is a cost and uptime tool. A company can send a simple task to a cheaper model and a harder task to a stronger one. That is the same function OpenRouter and others sell. The security layer is the added claim. Whether it holds depends on the attestation of the specific route and on the provider that ultimately runs the model. TrustedRouter’s own documentation notes that ordinary provider calls still send the prompt to the selected provider unless a stronger confidential-computing path is used. The $6 million is a seed. It is not a growth round. OpenRouter’s later sale set a high mark for the category. TrustedRouter is early. It has not published current daily token volume, revenue or customer count with this round. The earlier figure of more than one billion tokens in a day came with the August raise. Sierra Ventures led the round. The rest of the investor list was not published. A seed of this size can fund a small engineering team and more model integrations. It does not, by itself, prove that the confidential-computing routes are the ones customers use. The company will have to show that in traffic or in named customers. Perla has said encrypting AI chats can slow the cycle in which frontier labs absorb user data to train future models. That is a policy argument. It is not a measured reduction in data leaving a customer’s environment. A Gulf bank that already runs approved models inside a private or sovereign cloud does not need an external router for residency. A bank that wants many models and a documented encryption boundary does. The difference is the one the $6 million is meant to fund. If TrustedRouter publishes the share of traffic that runs inside the confidential-computing routes and a customer count, the security claim has a scale. If the next line is only another funding total, count the routes that actually encrypt end to end. The figure missing is both. That count is not yet public.