SAM #W911NF26RAIAI

Market Survey: Agentic AI Platforms for Army Classified Networks

The U.S. Army Combat Capabilities Development Command (DEVCOM) Army Research Laboratory (ARL), in partnership with Headquarters, Department of the Army, Military Intelligence, and the University of Arizona Applied Research Corporation (UA-ARC), is seeking information on Agentic AI platforms for Army intelligence use on classified networks. - Government Buyer: - U.S. Army DEVCOM ARL, Headquarters Department of the Army (Military Intelligence), and UA-ARC - OEMs and Vendors: - No specific OEMs or vendors are named; the RFI is open to all commercial, open-source, and Government-adapted solutions - Products/Services Requested: - Agentic AI building and management platforms for deployment on JWICS (priority), SIPRNet (IL6), and IL5 networks - Platforms must enable intelligence analysts to build, deploy, monitor, and govern AI agents capable of reasoning, information retrieval, and tool/function calling - Support for edge and disconnected, intermittent, or limited bandwidth (DIL) environments is desired - Respondents must provide details on product capabilities, accreditation status, deployment options, supported models, security controls, and pricing models - Stretch goal: platforms that can generate graph-based agents from natural language descriptions - Unique/Notable Requirements: - All responses must be unclassified - Proprietary information must be clearly marked - No specific products, part numbers, or quantities are listed - Solutions must be suitable for accredited classified environments

Description

The University of Arizona Applied Research Corporation (UA-ARC), in cooperation with Headquarters, Department of the Army, Military Intelligence, and U.S. Army Combat Capabilities Development Command (DEVCOM) Army Research Laboratory (ARL), is conducting a market survey to identify Agentic AI building and management platforms that can be deployed on Army classified networks. Priority is JWICS, followed by SIPRNet (IL6), then IL5.

The platform should enable Army intelligence analysts and staff to build, deploy, monitor, and govern AI agents capable of reasoning, retrieving information, and taking actions via tool and function calling inside accredited classified environments. We are also interested in solutions that support edge and disconnected, intermittent, or limited bandwidth (DIL) users who operate without persistent connectivity to a classified cloud or data center. Commercial, open-source, and Government-adapted solutions are all of interest.

This RFI is intended to survey industry capabilities and gather information on existing tools or platforms that meet the outlined requirements. Responses will inform future acquisition strategies, governance, and potential partnerships.

This RFI is open to both traditional and non-traditional large and small business concerns, government agencies, as well as academic and nonprofit entities to submit their technical feedback and capabilities that may meet the Army’s needs and objectives.

Disclaimer: THIS IS A REQUEST FOR INFORMATION (RFI) ONLY. This RFI is issued solely for information and planning purposes – it does not constitute a Request For Proposal (RFP) or a promise to issue an RFP in the future. This RFI does not commit the U.S. Government to contract for any supply or service whatsoever. Further, the Army is not at this time seeking proposals and will not accept unsolicited proposals. Respondents are advised that the Government will not pay for any information or administrative costs incurred in response to this RFI; all costs associated with responding to this RFI will be solely at the interested party’s expense. Not responding to this RFI does not preclude participation in any future RFP, if any is issued. Please note, the information received from this RFI is used to help the Government refine the requirement and help identify areas of ambiguity so that, if released, the RFP is well defined. Any feedback/questions as a result of this RFI may/may not be responded to directly and will be addressed in the final RFP if the Government deems it necessary.

Requested Information Please address the questions below. Be brief; bullet answers are welcome. Mark proprietary information clearly.

A. Company and Product (½ page max) 1. Company name, CAGE/UEI, business size, and point of contact 2. Product name, version, and a one-paragraph description 3. Notable Federal, DoD, or IC customers (redact if needed)

B. Accreditation Status 1. Current ATO status on JWICS, SIPRNet/IL6, and IL5. For each, identify the authorizing official, sponsoring organization, and boundary covered. Indicate whether authorization is current, in progress (with expected date), or not yet pursued 2. If not currently accredited for JWICS or SIPRNet, describe your realistic path and timeline to achieve accreditation, including any Government sponsor required 3. FedRAMP status (if applicable) and any additional relevant authorizations

C. Platform Capabilities 1. Agent authoring approach (no code, low code, code first) and supported agent patterns (single agent, multi agent, graph based) 2. Supported foundation models. Identify the country of origin, developer, and ownership status for each. 3. Tool/function calling, including allow-listing and human-in-the-loop controls for sensitive actions 4. Retrieval augmented generation (RAG) over enterprise data, including classification-aware access control 5. Identity integration (DoD PKI, ICAM), audit logging, and guardrails (prompt injection defenses, content filtering, policy enforcement)

D. Deployment (Classified Networks and Edge/DIL Environments) 1. Minimum viable footprint for a pilot on JWICS or SIPRNet (compute, GPU, storage, network, GFE assumptions) 2. How the platform handles software and model updates in air-gapped environments 3. Edge and DIL deployment options for users without persistent connectivity (e.g., tactical, deployed, or mobile users). Describe what agent capabilities are available locally versus those requiring connectivity, including supported hardware (laptop, ruggedized device, small server, embedded compute) and minimum specs (CPU/GPU, RAM, storage). 4. Smallest models the platform can run effectively on edge hardware, and how performance degrades relative to a full data center deployment 5. Synchronization model when edge users reconnect to the enterprise (agent state, retrieval indexes, audit logs, model and policy updates)

E. Pricing and Contract Vehicles (½ page max) 1. Pricing model (per user, per agent, consumption based, enterprise license, etc.) 2. ROM pricing for a 50-user pilot and a 5,000-user enterprise deployment 3. List available contract vehicles (GSA MAS, SEWP, CHESS/ITES-SW2, Tradewinds, OTA, CSO, etc.)

F. Implementation (½ page max) 1. Typical timeline from contract award to initial operational capability on a classified network 2. Cleared personnel availability for on-site support (TS/SCI where required)

G. Stretch Goal: Description-to-Graph Agent Generation Sections A–F describe the Government's baseline expectations. The capability described below is a stretch goal and is not required for a viable response. The Government is interested in platforms that allow a user to describe an agent in plain language and have the platform autonomously generate a working graph-based agent, without requiring the user to drag, drop, or manually configure individual nodes. 2 additional pages are authorized for addressing the stretch goal. Address the following: Current capability: Can your platform generate a complete, executable graph-based agent (nodes, edges, state schema, tool bindings) from a natural language description alone? If yes, describe the underlying approach (e.g., LLM code generation, meta agent, template synthesis) and provide examples of agents successfully generated this way. Validation and reliability: How does the platform verify that a generated agent is correct and safe before deployment (e.g., automated testing, simulation, evaluation harnesses, human review checkpoints)? What is the typical success rate of one-shot generation versus iterative refinement? Roadmap: If full description-to-graph generation is not currently available, describe your roadmap and expected timeline. Identify any research partnerships, ongoing pilots, or prototypes relevant to this capability.

Response Instructions: Provide a PDF or Word document, 10 pages maximum or 12 pages maximum if a stretch goal is addressed (not counting attachments). Attachments are limited to existing artifacts (e.g., ATO letters, FedRAMP package summaries, datasheets, architecture diagrams) and should not contain net-new narrative content. Links to short videos describing capabilities are welcome in the attachments. Classification: All responses must be UNCLASSIFIED. If Controlled Unclassified Information is available, provide a statement of relevance to help advisors determine if it should be requested using authorized methods. Do not send classified information.

Email submissions should be sent to the POCs below and be clearly marked as “Notice ID_Organization Name” in the subject line and on attachments. The Notice ID is displayed on the SAM.GOV notice.

Review Process: The Government will utilize cleared, U.S. citizen, non-government personnel to review and manage responses to this RFI and advise on specific technical and management matters and shall not, under any circumstances, be used as voting evaluators. However, the Government may consider the advice provided in its evaluation process. Non-government personnel will receive and assess email responses based on Government-approved criteria and operate under a non-disclosure agreement. All responses will be sent to the University of Arizona Applied Research Corporation (UA-ARC) points of contact, acting as a Partner Intermediary under 15 USC 3715.

Proprietary Information: The respondent must clearly state at the beginning of the response if a Proprietary Information Agreement (PIA), Non-Disclosure Agreement (NDA), or equivalent is required prior to non-government personnel reviewing the submissions. Should a PIA/NDA be required, the respondent must obtain the PIA/NDA from the UA-ARC POC below and provide the signed PIA/NDA with their RFI submittal. Respondents are responsible for adequately marking proprietary or competition sensitive information contained in their response. The front page of your response package should state "PROPRIETARY INFORMATION CONTAINED WITHIN", if applicable. All RFI submissions are treated as company proprietary information and the content will be disclosed to U.S. Government employees, military, or designated support contractors only for the purpose of this market research activity.

Contact Information: Primary Point of Contact: Philippe Bergeron pbergeron@ua-arc.org 520-626-3013 Alternate: Stephen Aldred saldred@ua-arc.org 520-621-4394 Note: Be advised UA-ARC is continuing business as usual while undergoing a formal name change to the Kyl Institute for National Security.

View original listing