Immigration and Customs Enforcement has been developing a new AI platform called STELLA that will give agency developers a secure, containerized environment to build and test large language model applications. The platform, detailed in documents obtained by FedScoop through a Freedom of Information Act request, represents a centralized push to standardize how ICE experiments with generative AI across its operations.
STELLA stands for Secure Technology Environment for Large Language Agents. According to the agency's response, the platform "will provide standardized toolsets, authentication/authorization, security/code scanning, AI guardrails, logging/analysis, and LLM access through the secure, FedRAMP ICE Cloud environment."
What the platform offers developers
The documents describe a system built for retrieval-augmented generation experiments, giving development teams access to containerized environments and the tools needed to prototype new AI use cases. ICE anticipates that once STELLA is deployed, additional use cases will be added as teams begin using the platform for Generative AI and LLM development.
The platform had an anticipated initial release this summer, though the actual deployment timeline remains unclear. In the Department of Homeland Security's AI use case inventory, updated in July, STELLA was still listed in a pre-deployment stage. ICE did not respond to requests for clarification on the schedule.
ICE did not disclose what third-party AI companies were involved in development, what data was used to train or fine-tune the tool, or the system's accuracy and hallucination rates. "The information requested is still being written, and the architecture and models to be used are still being finalized," the agency said.
Part of a larger AI portfolio
STELLA is one of more than 40 AI use cases ICE is currently developing or using. Others include mobile device forensics for investigations, open-source intelligence for lead identification, and tip processing. The platform is referred to as the ICE Enterprise AI Assistant in DHS inventory records.
Not all of ICE's AI projects have succeeded. The agency recently retired several use cases, including an AI-assisted resume screening tool and a system that evaluated whether noncitizens could remain in their communities while immigration cases were processed. The STELLA platform appears designed to give internal teams a more controlled environment for testing ideas before they reach production.
Why this matters for IT and development professionals
STELLA signals how government agencies are building internal AI development infrastructure rather than relying solely on vendor-built applications. For AI for IT & Development teams in regulated industries, the approach mirrors enterprise patterns: containerized environments, security scanning pipelines, guardrails, and centralized LLM access. The fact that ICE is still finalizing architecture and model choices also highlights how rapidly these internal platforms are evolving - and how much technical decision-making remains in flux even for large federal agencies.
Your membership also unlocks:
China bristles at Anthropic CEO's comments about its AI development