Don’t let AI liability be your agency’s next crisis.
Generic AI chatbots invent answers, leave non-English speakers behind, and produce interactions your records officer can’t audit. Agency Chat is the auditable alternative — grounded in your official sources, cited on every claim, captured for public records, and accessible in 72+ languages.
You need efficiency, but you can’t afford the risk.
Municipal leaders are under pressure to do more with less. But deploying the wrong AI creates problems far worse than the ones it solves.
The hallucination trap
NYC’s chatbot told businesses they could serve food with rat bites. Air Canada’s AI invented a refund policy — and the airline was held liable. The UNC School of Government warns: if your AI lies, you are legally responsible.
Equity debt is growing
GovLoop warns that AI tools routinely exclude people with disabilities and limited English proficiency. The federal Section 508 program now publishes a chatbot accessibility playbook because so many fail. The result is “silent attrition” — your most vulnerable residents simply stop trying to engage.
Civic isolation
Georgia Tech research found that generic chatbots make residents feel “isolated” — like their issues disappear into a black box. Bad automation replaces human connection instead of protecting it.
“It should be obvious to Air Canada that it is responsible for all the information on its website. It makes no difference whether the information comes from a static page or a chatbot.”
The UNC School of Government puts it bluntly: “If a local government employee had a pattern of producing incorrect, harmful, or biased content, that employee would likely face eventual discipline and dismissal. Is it prudent to invest in a generative AI chatbot that poses these same risks?”
Clone your best staff member — one who knows every policy, speaks 72 languages, and never sleeps.
Agencies that deploy grounded AI see dramatic results. Not because the AI replaces people, but because it frees your team to do the work that actually requires a human.
Calls reduced in 6 months by a London council using grounded AI.
Return on investment achieved in the London case study.
Average time from kickoff to live deployment with Agency Chat.
Grounded in your official sources.
Refuses when unsupported.
Agency Chat uses Retrieval Augmented Generation (RAG) — the architecture recommended by the UNC School of Government — to search your documents and website simultaneously. The assistant cites every claim and is designed to refuse when the answer isn’t supported by your official sources.
Document Intelligence — The 'Air Canada' Fix
Your municipal code, policies, and administrative regulations are indexed and searchable. Every response links directly to the source. If the answer isn’t supported by your official documents, the assistant says so rather than guessing.
The UNC School of Government recommends you “test out publicly available chatbots that the vendor has already built for other local governments.”
Website Search — Always Current
Automatically crawls your agency website for the latest information on services, schedules, and programs. Dual-source search means nothing falls through the cracks.
AI interactions are public records. We treat them that way.
Generic AI chatbots disappear into a black box. Agency Chat hands you the file. Every question, answer, and translation is captured to a dedicated agency-owned audit log, delivered monthly as a CPRA/FOIA-ready report, and then permanently purged from our systems on a fixed schedule.
Complete capture
Every interaction is logged with timestamp, language, the resident’s question, the assistant’s answer, and the citations used. Translations are captured alongside the originals — the full record of what your residents asked and what they were told.
Monthly delivery to your records officer
Every month, a structured report of every interaction is delivered directly to the addresses you specify — ready to archive against your existing public-records schedule. When a CPRA or FOIA request comes in, you already have the file.
Minimal retention, by policy
Records are kept on a two-month rolling window so a delivery can be re-issued if needed, then permanently purged from our systems. No PII is collected. No user accounts. No long-term archive sitting on a vendor’s servers waiting to leak.
See it run live.
Every Agency Chat deployment ships with a public transparency dashboard. Real traffic, real categories, real refusal rates — updated continuously from the same audit log that drives your monthly records report.
Why not just use a generic chatbot, or build one ourselves?
Generic AI is fast to ship and unsafe to deploy. A DIY RAG project on top of Vertex, Bedrock, or OpenAI can match the answer behavior — but the audit, equity, and operating layers are still on you. Agency Chat ships the whole stack as a managed service.
| Compare features | Generic AI | DIY RAG | Agency Chat |
|---|---|---|---|
| Answer behavior | |||
| Grounded in your official sources | Build it | ||
| Refuses when sources don’t support the answer | Build it | ||
| Citation on every claim | Build it | ||
| Multi-layer prompt-injection & adversarial guards | Best-effort | Build it | |
| Audit & public records | |||
| Every interaction captured to an audit log | Build it | ||
| Monthly CPRA/FOIA-ready records report | Build it | ||
| Public transparency dashboard | Build it | ||
| Minimal retention with scheduled purge | Build it | ||
| No PII collected, no user accounts | Varies | Build it | |
| Resident equity | |||
| 72+ languages with translated answers & citations | Some | Integrate | |
| WCAG 2.1 AA + Section 508 ready | Build it | ||
| Voice in & out (speech-to-text + text-to-speech) | Integrate | ||
| Operating model | |||
| Time to live deployment | Days (unsafe) | Months | 1–2 weeks |
| Document & website indexing managed for you | Build it | ||
| Capacity control & rate limiting | None | Build it | |
| Pricing model | Per-seat or token | Cloud + ops | Flat monthly |
| Answer behavior | |
|---|---|
| Grounded in your official sources | |
| Refuses when sources don’t support the answer | |
| Citation on every claim | |
| Multi-layer prompt-injection & adversarial guards | Best-effort |
| Audit & public records | |
| Every interaction captured to an audit log | |
| Monthly CPRA/FOIA-ready records report | |
| Public transparency dashboard | |
| Minimal retention with scheduled purge | |
| No PII collected, no user accounts | Varies |
| Resident equity | |
| 72+ languages with translated answers & citations | Some |
| WCAG 2.1 AA + Section 508 ready | |
| Voice in & out (speech-to-text + text-to-speech) | |
| Operating model | |
| Time to live deployment | Days (unsafe) |
| Document & website indexing managed for you | |
| Capacity control & rate limiting | None |
| Pricing model | Per-seat or token |
| Answer behavior | |
|---|---|
| Grounded in your official sources | Build it |
| Refuses when sources don’t support the answer | Build it |
| Citation on every claim | Build it |
| Multi-layer prompt-injection & adversarial guards | Build it |
| Audit & public records | |
| Every interaction captured to an audit log | Build it |
| Monthly CPRA/FOIA-ready records report | Build it |
| Public transparency dashboard | Build it |
| Minimal retention with scheduled purge | Build it |
| No PII collected, no user accounts | Build it |
| Resident equity | |
| 72+ languages with translated answers & citations | Integrate |
| WCAG 2.1 AA + Section 508 ready | Build it |
| Voice in & out (speech-to-text + text-to-speech) | Integrate |
| Operating model | |
| Time to live deployment | Months |
| Document & website indexing managed for you | Build it |
| Capacity control & rate limiting | Build it |
| Pricing model | Cloud + ops |
| Answer behavior | |
|---|---|
| Grounded in your official sources | |
| Refuses when sources don’t support the answer | |
| Citation on every claim | |
| Multi-layer prompt-injection & adversarial guards | |
| Audit & public records | |
| Every interaction captured to an audit log | |
| Monthly CPRA/FOIA-ready records report | |
| Public transparency dashboard | |
| Minimal retention with scheduled purge | |
| No PII collected, no user accounts | |
| Resident equity | |
| 72+ languages with translated answers & citations | |
| WCAG 2.1 AA + Section 508 ready | |
| Voice in & out (speech-to-text + text-to-speech) | |
| Operating model | |
| Time to live deployment | 1–2 weeks |
| Document & website indexing managed for you | |
| Capacity control & rate limiting | |
| Pricing model | Flat monthly |
Built to pay down equity debt, not create it.
University of Nebraska research found that text-only interfaces fail elders. GovLoop warns that inaccessible AI creates “silent attrition.” Agency Chat is designed from the ground up to serve everyone.
72+ Languages
Automatic translation for Spanish, Chinese, Vietnamese, and dozens more. Questions are translated for search, answers delivered in the resident’s language.
Voice In & Out
Speech-to-text input and text-to-speech output. Seniors and low-vision users can speak their question and hear the answer read aloud.
WCAG 2.1 AA
Keyboard navigation, screen reader support, and high contrast mode. Section 508 compliance built in, not bolted on.
Privacy by Design
Built to help government agencies meet public records requirements. Monthly reports deliver each interaction's date, time, question, and response directly to the agency. No Personally Identifiable Information is collected, and all chat records are permanently deleted after delivery.
Frequently Asked Questions
Air Canada’s chatbot invented a refund policy that didn’t exist, and a tribunal held the airline liable. NYC’s chatbot told business owners they could serve food with rat bites and violate labor laws. The UNC School of Government cites these cases as warnings: if your AI gives bad advice, your agency bears the legal responsibility.
The UNC School of Government recommends that you “test out publicly available chatbots that the vendor has already built for other local governments.” We agree. Try our live City of Oakland deployment at askoakland.ai right now — ask it anything about Oakland’s municipal code, zoning, or city services. Then check the public transparency dashboard to see real traffic and refusal behavior. For the architecture, safety, and records-handling detail your CIO or city attorney will want, see our security & architecture overview.
Every interaction is captured to a dedicated agency-owned audit log and delivered monthly to your records officer as a CPRA/FOIA-ready report — date, time, language, the resident’s question, the assistant’s answer, and the citations used. We retain records on a two-month rolling window so a delivery can be re-issued if needed, then permanently purge them. No PII is collected, and no user accounts are required. See the security & architecture overview for the full data-handling lifecycle.
Yes. Agency Chat is built with WCAG 2.1 AA compliance, including keyboard navigation, screen reader support, high contrast mode, and text-to-speech for answers. This directly addresses the “Equity Debt” that GovLoop warns about — where inaccessible digital tools exclude the most vulnerable residents.
The opposite. Georgia Tech researchers found the key is design: by automating routine questions (trash schedules, parking rules, permit requirements), your staff is freed to spend time on complex issues where human empathy and judgment matter most. Agency Chat handles the routine so your team can build better community relationships.
Costs less than one part-time staff role — and works 24/7.
A single department typically spends $60k–$100k per year answering routine questions. Agency Chat is a fraction of that, with no lock-ins, no per-seat licensing, and no hidden fees.
Base Subscription
$1,495/month
Fully functional site that shows its work.
- 10,000 AI interactions per month
- 5M DeepL or 6M Google translated characters
- Full document & website search
- AI answers with citations
- Your own subdomain
No enterprise drama
No multi-year lock-ins, no per-seat licensing, no hidden implementation fees, no vendor negotiation theater. Cancel anytime.
Need more capacity?
Additional AI interactions from $17.50/1,000 (prepaid) or $25/1,000. Translation from $25–$30/1M chars (prepaid) or $30–$40/1M chars.
Usage is metered monthly. Prepaid credits are consumed first, then overages are billed in arrears.
Governance, risk, and the AI frontier
Apr 20, 2026
Procurement as Risk Architecture: Why Pricing Structure Is a Governance Decision
April 20, 2026 — When procurement structure becomes governance risk: the accountability gap inside public-sector pricing architecture…
Read more
Apr 15, 2026
Procurement Is the Real AI Strategy
AI governance and the procurement gap: how contract terms determine institutional control — April 13, 2026…
Read more
Apr 8, 2026
The Best Chatbot for Cities and Municipalities: Why Agency Chat Is the Strongest Institutional Fit
Why Governance, Liability, and Procurement Realities Make Agency Chat the Safest AI Assistant for Municipal Governments…
Read more
Stop worrying about liability. Start serving your community 24/7.
See Agency Chat in action right now with the City of Oakland. Then start your 30-day live evaluation.