Thread AI and US Army develop fire-control research tools

Thread AI has entered a Cooperative Research and Development Agreement (CRADA) with the US Army’s DEVCOM Armaments Center to develop an artificial intelligence application for engineers and scientists working on fire-control systems. The collaboration will use Thread AI’s Lemma orchestration platform to connect specialised technical information and support research activities where access controls, source traceability and the reliability of generated answers are important. The agreement establishes a development programme rather than the acquisition of an operational fire-control system.

The proposed application will bring together technical resources accumulated through weapons systems research and development, allowing authorised personnel to examine information held across different sources. Thread AI says conventional general-purpose AI tools are limited when dealing with specialised engineering material that falls outside their original training data. Its approach centres on connecting an AI model to the relevant technical records and controlling how those sources are accessed, processed and referenced during research. The objective is to reduce the time required to locate and synthesise information without removing the need for engineering judgement.

Researching fire-control systems requires engineers to relate ballistics, sensing, software controls and weapon interfaces while checking which source records apply to the configuration under study. Technical knowledge may be distributed across design documentation, research records, test reports and software development material, with individual sources describing different configurations or stages of a programme. An engineer examining a particular technical issue may need to establish which records apply to the equipment under consideration before drawing conclusions from them. Connecting these sources electronically could make that process more efficient, provided the application preserves their context and identifies the evidence supporting its output.

Thread AI’s Lemma platform will provide the orchestration layer for the proposed research application, connecting the approved technical resources and software tools. Rather than depending exclusively on information held within a model, the application is intended to retrieve relevant material from connected resources and use it during the reasoning process. The design also provides mechanisms for recording the steps undertaken by a workflow, allowing its operation to be examined after a result has been produced.

Where technical documents describe successive equipment revisions, an AI-assisted answer must preserve the distinction between those configurations and the evidence supporting each claim. A specification may apply to one configuration, while a subsequent revision records an altered component, calculation method or operating limit. Combining information from both sources without recognising their differences could produce an answer that appears plausible but does not correspond to either approved configuration. Source traceability allows researchers to inspect the material used by the application and assess whether it supports the conclusions presented. Technical correctness still requires independent engineering assessment.

Thread AI intends to address these requirements through Lemma’s governance and workflow controls, which can determine how different components obtain access to information and record the actions they perform. The platform supports configurable permissions, integration with external software and checkpoints requiring human involvement. Such controls are relevant where researchers work with information that cannot be made available indiscriminately across an organisation. The specific access arrangements, deployment environment and technical security accreditation for the DEVCOM application have not been disclosed, however, so no particular classified information system can be presumed to have been integrated.

According to the company’s product architecture, a Lemma workflow consists of defined processing steps that can combine conventional software operations with AI-assisted reasoning and authorised human decisions. Each execution maintains information about its progress, while the platform is designed to preserve records of the operations performed. Applied to engineering research, this architecture could allow a user to submit a technical query, retrieve authorised source material, examine the resulting analysis and inspect the records supporting it. The exact workflow selected for the Army project will be determined through development and testing rather than inferred from the platform’s general capabilities.

The DEVCOM Armaments Center already develops and maintains safety-critical software across several weapons programmes, including applications concerned with ballistic calculations and artillery fire control. Its wider modernisation activity includes the United States Ballistics Library, Artillery Execution Suite and Common Fire Control Framework, alongside efforts to improve software development and integration. These existing programmes demonstrate the specialist engineering environment in which the new research application is being developed. Thread AI has not announced integration with any of those named systems, and the research agreement does not identify an operational weapon platform that will receive Lemma.

Within the Armaments Center, digital engineering initiatives are also intended to improve the exchange of information between tools that were previously managed separately. Bringing technical data into a more consistent environment can reduce manual transfer and help teams work from compatible records, although the engineering authority responsible for approving a design or calculation remains unchanged. The proposed Thread AI application will address a related information-management requirement by helping researchers work across specialist sources. Its effectiveness will depend on the quality of the connected material, the accuracy of retrieval and the ability of engineers to inspect the resulting evidence.

A cooperative research agreement provides a framework for government and industry specialists to undertake mutually defined development activities, with each organisation contributing appropriate expertise or resources. It differs from a conventional production procurement and does not automatically establish a commitment to buy a completed system. In this case, the Army brings fire-control research expertise and knowledge of the technical environment, while Thread AI supplies the orchestration platform and associated engineering capability. The companies have not disclosed a financial value, duration, test schedule or detailed allocation of development responsibilities.

The Army will still need to assess the application’s ability to retrieve the correct records and interpret them accurately before any wider adoption. A system may retrieve documents correctly while still generating an incomplete or incorrect interpretation of their contents. Establishing confidence in a research tool therefore requires assessment of how it handles source selection, conflicting information, missing records and questions that cannot be answered from the available evidence. DEVCOM’s evaluation methodology and accuracy thresholds remain undisclosed, and no completed validation results have been released.

Thread AI has previously deployed Lemma within commercial organisations and describes the Army collaboration as a formal expansion of its federal and defence activities. Its existing software infrastructure provides a starting point for the project, but the specialist fire-control research environment introduces additional requirements that must be addressed through the collaboration. The parties will develop an application intended to make technical information more accessible while maintaining controlled access and traceable outputs. Further milestones will depend on implementation, testing and the Army’s assessment of the application, with any subsequent operational adoption requiring separate decisions.