Consus launches AI gateway for ITAR and CUI-compliant model access

Writing in a Launch HN post, Eric said he spent most of his career before Consus in aerospace and defense, launching national security payloads at SpaceX and joining Castelion as an early employee. He said the sector is growing quickly but that most companies struggle to use AI because of data protection requirements and compliance risk, citing a satellite manufacturer he worked with that does not allow any AI tools, even for non-sensitive work. He said he was surprised by how many aerospace and defense companies refused to touch AI despite seeing its usefulness, because they feared model providers training on their data, violating an ITAR rule, or losing business after a failed cyber assessment.

Consus is built as a compliant AI inference layer that plugs into existing tools and workflows, the company said. It can connect to tools such as ChatGPT or Claude, open-source options including the Pi coding agent and Open WebUI, and applications maintained by individual organizations. Every request reaches a single OpenAI-compatible endpoint where keys are verified and policy is enforced before anything touches inference, according to the post. Model names must carry a compliance suffix, such as gpt-5.6-sol:fedramp-high or claude-opus-5-5:itar, and bare model names are rejected. Eric said Consus never proxies to standard commercial endpoints.

Customers can also build an allowlist of models and assign spend and model restrictions to individual users, managed through a Consus Portal that handles spend, keys and policies, the company said.

Addressing why customers would not use Bedrock in AWS GovCloud or Azure Government directly, Eric said access to those platforms is difficult for small teams, while larger organizations inherit some but not all of the controls needed for sensitive data. Models are also split across providers, he said: GPT-5.6 Sol is available only in Azure Government, while Claude Opus 5.5 is served through Bedrock in AWS GovCloud, meaning customers who want both must stand up and secure two separate government cloud environments. Consus runs them behind one endpoint.

Given the sensitivity of the data, Eric said Consus stores no customer data at rest. Prompts and completions stay in memory only for the length of the request and are never logged or written to disk, with zero data retention enforced on Consus and on the infrastructure and model providers, he said. Metadata on each request and response is kept for billing and audit purposes.

Onboarding is not self-serve by design. Eric said ITAR requires the company to know exactly who is behind every key, so each organization is verified and a contract put in place before API keys are issued. Consus charges a compliance fee starting at $40 per month per seat plus AI tokens at cost, he said, adding that the company is actively serving customers and inviting feedback from security experts and engineers in regulated industries. The post included a demo video link and documentation for its API, security architecture and compliance details.

The Launch HN post had received 4 points and 1 comment on Hacker News. Details about the compliance audits, customer traction and pricing come from the company's own announcement.

Editor's Summary

Consus launched an AI gateway that routes model requests only to infrastructure it says meets ITAR and controlled unclassified information requirements, after completing a FedRAMP High equivalency audit. The service targets aerospace and defense firms that have avoided AI tools over compliance and data-training concerns, and charges $40 per seat per month plus tokens at cost. Claims about audit status and customers come from the company's own launch post.