The Australian Government has issued a Request for Expressions of Interest (REOI) relating to the future expansion of GovAI, including GovAI chat and other associated capabilities to support the continued evolution of shared AI services across government.
GovAI is a whole-of-government AI service led by the Department of Finance designed to support the safe, secure and responsible adoption of AI across the Australian Public Service (APS).
GovAI aims to build APS capability by providing shared training, common tools, secure infrastructure, and practical AI solutions that allow agencies to safely learn, trial and apply generative AI.
To date, these include GovAI Chat, the Interactive Learning Environment, APS AI Library, Model Brokerage Service, and AI hosting service solutions.
The REOI relates to stage one of an intended multi-stage procurement process, with Finance currently seeking to better understand market capability, maturity and potential suitable products and services, along with any indicative commercial approaches; no contract will be awarded, and no commitment of funding will be made, as a result of this first stage of the REOI.
Stage two, if required in the future, will involve one or more targeted approaches to market with selected respondents from stage one to progress specific work packages, subject to government decisions and future funding, and may include targeted approaches to vendors.
GovAI Chat to be evaluated
A key part of this first stage is seeking expressions of interest from respondents able to provide a potential GovAI Chat alternative.
GovAI Chat is a generative AI assistant for APS staff providing central access to AI models, designed as a secure environment with onshore processing, APS-controlled data governance and auditable responses and integrated training.
The service provides a web-based conversational interface to support everyday public sector work, including drafting documents, analysing content, interpreting policies and generating ideas. It also supports file uploads, enabling users to query documents directly and undertake more contextual workflows.
The new app being sought must include modular features, extensions or components that could replace the secure, whole-of-government AI chat client for approximately 200,000 users.
Essential requirements and capabilities include a multi-vendor and multi-model capability to avoid vendor lock-in and providing model choice for the user depending on their needs, including possible dynamic model switching based on user prompts.
It needs to be hosted entirely onshore, including data storage, transmission and model processing, along with adhering to controls from the Australian Government’s Information Security Manual and Protective Security Policy Framework, architected and hosted to pass IRAP certification to PROTECTED.
The solution must be authenticated via the Australian Government’s Single Sign-on capability, VANguard; include networking controls to restrict access to selected government networks; feature administrative controls for configuration, entitlements, monitoring and auditing, including enforcing budget management/token capping to manage costs at an individual user level; and include multiple levels of safety, assurance and guardrail features supporting responsible AI use.
Beta phase expected in near future
GovAI Chat is currently in an Alpha phase, which commenced in April 2026, as a minimum viable product undergoing real-world testing. The service presently supports up to 5,000 opt-in users working with OFFICIAL-classified information and is accessed via both GovTEAMS credentials and secured government devices through GovLINK.
A broader Beta phase is planned, with increased user numbers, enhanced functionality and the use of agency credential Single Sign-on via VANguard. The intention is that the app will support PROTECTED data from late 2026.
Expanding access to GovAI and GovAI Chat was a key pillar of the AI Plan for the APS introduced last year, which aimed to improve government service delivery, policy outcomes, efficiency, and productivity, through substantially increasing the use of AI in government.
GovAI Chat currently uses a multi-model architecture that brokers access to different external AI providers, allowing users to choose the model according to task suitability.
Data is routed through onshore deployments in Australian data centres so that it remains within national jurisdiction and the platform combines a user-facing interface with brokerage and orchestration controls, and cloud infrastructure hosted across Azure and AWS in line with government security requirements.
Security, safety and governance are embedded throughout the platform design, operating across multiple layers, including procurement arrangements, infrastructure configuration, model-specific behavioural constraints and system prompts aligned with Australian Government policy.
This approach is guided by the GovAI Chat Safe Responsible Design Framework and supported by relevant government standards and policy settings. All data is stored and processed in Australian-certified hosting environments, with no offshore transfer, and the use of personal information is restricted during the Alpha phase.
Governance operates on a shared responsibility model and the central GovAI team manages platform capability and security, while participating agencies remain accountable for staff compliance and internal policy adherence.
MLOps, secure integration and cloud data platforms also sought
Further to the requirements around GovAI Chat, Finance is also seeking EOIs from respondents able to provide products and services that may be relevant to either integrate with GovAI Chat or as standalone capabilities for use by Finance/GovAI and/or other APS agencies.
The first of these is around AI platforms that enable the management and federation of models, including frontier models and hosted open-source models.
Any proposed solution should enhance a broad range of model classes, including language (both Large Language and Small Language), multimodal, embedding, vision, speech, generative media, predictive, graph, reinforcement learning, and agentic models, enabling flexible selection based on task and use case.
Respondents able to provide tools or software for cloud hosting of Model Context Protocol (MCP) Gateways, hosting platforms, and integration between systems for use with AI models and AI supported solutions, along with platforms that provide a secure, governed environment for the storage, discovery, sharing, reuse and lifecycle management of data assets, are also sought.
For the data asset platform, Finance has advised that the platform should operate as a trusted data asset repository, enabling structured and unstructured data to be hosted, governed, versioned, accessed and shared across organisational boundaries, including federate secure data sharing solutions.
Reusable assets for APS AI Library needed
GovAI is also currently developing the APS AI Library, which is a whole-of-APS library to share AI use cases, templates and reusable assets.
To support GovAI in building reusable assets for the APS AI Library, Finance is seeking expressions of interest from respondents able to provide modular AI application templates.
These application templates are intended to serve as configurable patterns for common use cases, allowing APS agencies to streamline and accelerate their AI projects and ideas, while tailoring them to their specific operational contexts.
Finance has completed extensive consultation and co-design activities to identify a potential future suite of reusable concepts and modular AI ideas, based on their perceived value to government.
As such, potential respondents need to be able to provide either existing commercial-off-the-shelf solutions and/or proposals for the development of applications that meet a number of proposed use cases.
These use cases include drawing on the APS style guide to support content review and editing; summarising consultation responses and identify key themes, insights and recommended actions; reviewing large documents against set criteria; and optimising government web content to support accurate answers from 3rd party AI models.
They also include the ability to compare build, buy, or borrow procurement paths for government AI procurements; combine DTA and agency policy guidance to help users understand AI use case risks and required controls; develop policy responses using a RAG application for specific policy areas; flag and redact sensitive content while prepping material for RAG use; and turn standard operating procedures into permissions-aware, cited answers staff can trust.
The closing date for the REOI is 2:00pm (ACT local time) on Friday 24 July.
.jpg&h=142&w=230&c=1&s=1)
.jpeg&h=142&w=230&c=1&s=1)
.jpg&w=100&c=1&s=0){kind=logo}
.jpg&w=100&c=1&s=0)
.jpg&w=100&c=1&s=0)
.jpg&w=100&c=1&s=0)
