Vendor Madison AI demos county knowledge assistant; commissioners discuss security, cost and timeline

A representative of Madison AI presented a live demonstration of a county-specific “knowledge assistant” to Rio Blanco County commissioners at the July 8 work session, and commissioners discussed possible deployment, cost, data ownership and IT-security concerns.

Hayden, identified as head of marketing and a member of Madison AI’s founding team, described how the company ingests a jurisdiction’s institutional records (agendas, minutes, ordinances, budgets and other public documents) into a closed dataset and builds tailored AI “agents” to speed routine government research tasks. “What we do is we build AI assistance to accelerate sort of the everyday work for managers, clerks, electeds, and department leaders,” Hayden said during the presentation.

Madison AI proposed a roughly 60-day deployment timeline and quoted an annual subscription price of $12,500 for Rio Blanco County’s size; the company said that would include unlimited users and an operational usage allowance (they described a default monthly allowance of 20,000 messages). Hayden said the system would be hosted on Microsoft Azure in the United States and that the company does not train other models on customers’ private data. “All of the data, all of the data acquisition, all of the data storage, and all of the AI that we're building is in The United States, and it's hosted on United States servers,” Hayden said.

Commissioners and staff pressed for details on data ownership, security, user counts and ongoing updates. Madison AI told the board the county would own its data, that the company would provide copies and that a contractual “nuclear” deletion option would permanently remove the deployment if the county terminated service. The company also said it can update its index weekly and can support automated ingestion from Google Drive or other repositories; small file sets (20–30 documents) could be processed in hours while larger datasets require days.

The discussion included operational use cases Madison AI demonstrated: finding prior ordinances and resolutions, compiling historical voting and fiscal-impact timelines for a topic, and drafting, then iterating, letters of support or opposition using uploaded legislative text. Commissioners asked whether the assistant hallucinated answers; Hayden responded that Madison AI is tuned to reduce hallucinations and that the system will say when requested information is not present in the county’s indexed documents.

Commissioners flagged several procurement and funding questions. Manager Edwards noted the county could pursue DOLA or AGNC grant funding; Madison AI said DOLA grants have been used elsewhere to cover roughly 50% of similar deployments. Commissioners also requested technical security documentation and said the county’s IT staff must review vendor specs before a decision. One commissioner asked whether data would remain accessible to the county if the contract ended; Madison AI responded that the data and prompts belong to the county and could be exported or destroyed on termination.

No formal direction to proceed was recorded at the work session; commissioners asked staff and the Madison AI representative to follow up with IT, provide security and deployment specs, and explore grant funding and a pilot deployment timeline that could deliver a usable county dataset in weeks rather than months.