Onshore Intelligence Pre-launch

Ask your own documents. Keep them inside your network.

Onshore Intelligence is building a local SI agent for engineering teams whose data cannot go to the cloud. It answers from your own documents, shows the source behind every line, and runs on open models installed on hardware you control.

Demo data: a fictional company and synthetic documents.

Cited, line by line

Every sentence in an answer points to the passage it came from. Hover to preview it. Click to open the document at that section or page.

Runs where your data lives

Documents, search index and model stay on your hardware. The models are open-weight releases from U.S. developers, so nothing depends on an outside service.

Tested on your documents

Each candidate model is tested on questions drawn from your own documents and scored for accuracy, speed and cost before anyone relies on it.

One agent for everything your team has written down.

Specifications, interface documents, test reports, anomaly reports, procedures, standards. Ask in plain English and get an answer you can trace to its source.

  • ChatAsk follow-up questions. The agent keeps the context of the conversation.
  • CitationsEach sentence carries the passages it rests on. Hover to preview a passage, click to open the document at that section or page.
  • No guessingAnswers are written only from retrieved passages. When the documents do not contain the answer, the agent says so.
  • LibraryPoint it at a folder or upload files. PDF, Markdown and text today. Each file is indexed on your own machine.
  • ModelsOpen-weight models such as Gemma, Llama, gpt-oss and Granite. Swap one for another without moving a single document.
  • Coming nextConnectors to document systems, per-user permissions, and a coding assistant for teams that cannot use cloud coding tools.

Your designs, test data and controlled files never leave your network.

Export rules treat two things as an export of controlled technical data: sending it out of the United States, and releasing it to a foreign person, even inside the country. A cloud SI service puts another company, its staff and its servers in the path of that data.

Everything runs on hardware you control.

The documents, the search index and the model weights sit on a server inside your network. No outside service is called to answer a question, and no vendor staff can see what was asked.

The models are open-weight, so you hold the weights. Nobody can retire the model you validated or change it under you.

Who inside your company may see which documents remains your decision.

What stays outside
  • Cloud SI servicesNot called.
  • Third-party APIsNot called.
  • Servers abroadNothing is sent or stored outside your network.
  • Vendor staffNo remote access to your documents or questions.
  • Model trainingYour files are never used to train anyone's model.

Every model is proven on your own documents first.

Before your team relies on a model, the product measures it on questions drawn from your own documents.

  • Your questionsThe product writes test questions with reference answers from your material, including questions whose answer is not there.
  • Same test for allEach candidate model answers from the same retrieved passages.
  • Three gradesIs the answer correct? Does each cited passage support its sentence? Does the model decline when the answer is absent?
  • Speed and costReported next to quality, so you can choose a model and re-run the test when a new one is released.

Built for teams that work under export controls.

Aerospace, space and defense suppliers handle technical data that may not be sent abroad or released to foreign persons. Onshore Intelligence is being built for them.

  • StageWorking prototype. Not yet available for purchase.
  • Based inUnited States.