Open Source · Built from Inside City Hall

The AI harness for local government.

CivicWork is building CivicAide — open-source infrastructure that connects AI models to municipal data, workflows, and governance guardrails. Not a vendor. Not DIY. A partner for municipalities ready to own their AI infrastructure. We're pre-deployment, working with our first municipal partner through U.S. Digital Response, and looking for local governments ready to collaborate on what comes next.

Learn About CivicAide →View on GitHub

Where We're Focused

The Challenge

A capacity crisis, a data problem, and a vendor dependency that nobody talks about.

The same planner reviewing a 47-page environmental impact study is also answering phones, attending three committee meetings, and trying to update a comprehensive plan that's been outdated since 2018. The expertise is in the building. The hours aren't.

Meanwhile, the data that should power better decisions sits scattered across systems that don't talk to each other — half of it in PDFs from 2011. Vendors build products around this public data and charge municipalities to access their own information. A mid-size city pays $14,000 a year for public records request software — a system that is, at its core, a form, a queue, and a deadline tracker. Twenty thousand municipalities can't afford even that, and manage a legally mandated process through email and spreadsheets.

Now agents are coming. And most local governments aren't ready — not because they lack ambition, but because their data layer wasn't built for it.

CivicWork is building tools designed to give existing staff more capacity. We don't automate people out of jobs — we automate the tedious parts of jobs so qualified professionals can spend their time on the work that actually requires their expertise.

Capacity gapsFragmented data systemsVendor lock-in on public dataNo path to agent-readiness

The Platform

CivicAide: the harness that makes AI useful for local government.

AI models are powerful but generic. CivicAide is the harness — the data connections, workflow integration, and governance guardrails that make the engine useful for public service. It started as a Coda doc for one part-time councilmember. As frontier models and agent frameworks matured, so did the ambition. Not everything here is AI. Automation is sufficient where it fits.

In testing

PolicyAide

Multi-agent policy analysis inspired by Google’s CoScientist paper. Specialized agents pressure-test proposals from multiple angles before they reach a decision-maker. Built on Anthropic’s Agent SDK. Far along — not yet public.

Building now

Data Layer & Agent Readiness

Designed to help municipalities normalize existing data, process current workflows with open-source tools, and prepare their infrastructure for agents. The shift from human-centered design to agent-centered design starts here.

Contributing

Open-Source Tooling

The Municipal Plugin and Coda MCP Server are live and open source. WebMCP integrations for municipal platforms are shipping soon. In development: a vendor contract analysis workflow that helps municipalities see what they’re paying for and what could be built with open tools.

Open SourceAuditable · Explainable · Designed for public scrutiny

Explore CivicAide in depth →

Frameworks for Saying No

Most AI vendors sell what's possible. We start with what's appropriate.

Before building any tool, we ask two questions: Can we verify that it works? And does deploying it build or erode the trust a community needs? Together, the Verifiability Framework and Trust Stack form a diagnostic system for responsible municipal AI.

Framework 01

The Verifiability Framework

One question: Can we verify this? If you can define success criteria and automatically evaluate outcomes, the task is an automation candidate. If you can't, AI should augment human judgment, never replace it. And when a use case is mixed — decompose it, automate what you can, keep humans on the rest.

Verifiable → Automate

Permit completeness checks, document classification, data validation, compliance verification. Clear criteria, fast iteration, measurable outcomes.

Mixed → Decompose

311 triage, public chatbots, hiring, inspection scheduling. Separate the automatable components from the parts requiring human judgment.

Non-verifiable → Augment only

Budget priorities, policy tradeoffs, community needs, accountability decisions. AI expands what you know. Humans retain authority over who decides.

Read the full framework →

Framework 02

The Trust Stack

Four layers of trust, each building on the last. Layers 1–2 are built with technical tools — audits, testing, metrics. Layers 3–4 require democratic process — deliberation, community input, accountability. The common failure: trying to solve Layer 3–4 problems with Layer 1–2 tools.

Process TrustTechnical

Did it follow the rules?

Outcome TrustTechnical

Did it work?

Representation TrustDemocratic

Does it reflect us?

Sovereignty TrustDemocratic

Do we still control this?

Read the full Trust Stack →

“AI should expand what you know, not replace who decides.”

These frameworks give municipal leaders structured language for evaluating AI proposals — and principled grounds for declining when a deployment doesn't fit, without sounding like they're against technology.

Currently partnering with U.S. Digital Response (USDR) on AI use case discovery across Elgin city departments — using the Verifiability Framework and Trust Stack as diagnostic tools to identify where AI helps and where it doesn't.

Built From the Inside

“I started building CivicAide because I needed it — a part-time councilmember with a day job, trying to get through 200-page packets before the next vote. It was a Coda doc connected to a few LLMs. Then frontier models got good enough to do real policy analysis. Then agent frameworks matured enough to orchestrate it. The tools kept getting better, so the ambition kept growing.”

Dustin Good is a sitting at-large city councilmember in Elgin, Illinois — second term, 4.5 years in office. CivicWork grew out of the problems he encountered firsthand: too much data, too little time, and a growing conviction that local government shouldn't have to wait for vendors to solve problems practitioners already understand.

4.5 yearsin elected office

2 termsas at-large councilmember

SpeakerIML & City AI Connect conferences

Our Approach

How we think.

Practitioners first

This wasn’t designed from a research lab or a consulting firm. It was built from inside a municipal building — from the meetings, the packets, and the cross-departmental workflows that make local government actually function. The tools reflect actual workflow, not a theory of it.

Automation before AI

Not everything needs a language model. Sometimes a well-structured automation does the job. Just as important: CivicWork helps municipalities say no to AI where it doesn’t belong. Staff are getting pitched AI solutions for everything — having a principled framework for declining, not out of fear but out of clarity, is as valuable as knowing where to deploy.

Own the data layer

When a municipality wants to leave a software vendor, they often discover their own data — years of records, workflows, and institutional knowledge — can’t come with them. Agent-readiness means making public data actually accessible, portable, and structured so AI tools can help staff do their jobs. That’s not a technology problem — it’s a data ownership problem, and it’s what we’re building toward.

Open source, open trust

Open source doesn’t mean less secure — the world’s most security-critical infrastructure runs on open-source software. For public institutions, open source means inspectable, auditable, and not dependent on a vendor’s continued existence. Government AI tools should be verifiable by the people they serve, not black boxes controlled by vendors.

Let's build together.

CivicWork is looking for municipal partners willing to be early collaborators — local governments ready to explore what governance-grade AI infrastructure looks like in practice. If you're a municipality, a civic technologist, or building in this space, let's talk.

Get in Touch View on GitHub →