Consus launches FedRAMP High LLM gateway for ITAR data — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Consus (YC F24) routes AI requests to ITAR-restricted and CUI-compliant infrastructure, having completed a FedRAMP High equivalency audit against all 410 controls — the U.S. government's strictest cloud security baseline for cloud services handling sensitive unclassified data.
- Eric, the founder, previously launched national security payloads at SpaceX and was an early employee at Castelion, and built the gateway after watching aerospace and defense firms refuse AI over ITAR and compliance fears — including a satellite manufacturer that bans AI even for non-sensitive work.
- Every model request must carry a compliance suffix (e.g.,
gpt-5.6-sol:fedramp-highorclaude-opus-5.5:itar) on an OpenAI-compatible endpoint; bare model names are rejected and the system never proxies to standard commercial endpoints. - Consus runs across AWS GovCloud, Azure Government, and Google Cloud, letting customers access models split across providers through one endpoint — GPT-5.6 Sol on Azure Government, Claude Opus 5.5 on AWS GovCloud — without standing up separate government cloud environments for each.
- The platform enforces zero data retention — prompts and completions stay in memory only for the duration of a request and are never logged to disk — verifies every organization and signs a contract before issuing API keys (ITAR requires it), and charges $40 per seat per month plus AI tokens at cost.
- The gateway connects to ChatGPT and Claude on the commercial side, open-source tools like the Pi coding agent and Open WebUI, or custom applications — letting customers plug a single compliance layer into any AI workflow, just as OpenRouter does for non-regulated teams.
Why it matters: For aerospace and defense firms locked out of commercial AI tools by ITAR and CUI rules, Consus turns a multi-GovCloud stand-up into a single $40/seat endpoint — letting ITAR-restricted teams use frontier models like Claude Opus 5.5 and GPT-5.6 Sol that previously required standing up two separate government cloud environments.
Ask SkimNews
