Agentic Government Oversight: the systematic, automated exercise of transparency rights the public already has.
Every public employee in America is already legally observable. FOIA, state public records acts, open meetings laws, body camera mandates, financial disclosure requirements, court records -- the infrastructure exists. Nobody uses it at scale because humans don't have the bandwidth.
AI agents delete that excuse.
This toolkit lets you deploy automated oversight agents against any U.S. city or county government using only publicly available data and existing legal frameworks. No new laws required. No hacking. No surveillance. Just the systematic exercise of rights you already have.
Cost: ~$4.12 per public employee per year. $38/month covers a city of 80,000.
📖 Read the full analysis: Every Public Employee in America Is Already Legally Observable. AI Agents Could Actually Do It.
agent-watchdog/
├── config/ # City configuration files
│ ├── city-template.yaml # Template -- fork and fill in for your city
│ └── example-menlo-park.yaml # Working example: Menlo Park, CA
├── modules/
│ ├── foia-engine/ # Automated FOIA/PRA request generation & tracking
│ ├── meeting-monitor/ # Public meeting transcript ingestion & analysis
│ ├── financial-disclosure/ # Cross-reference disclosures with policy votes
│ ├── permit-inspector/ # Building permit & inspection anomaly detection
│ └── dashboard/ # Static public dashboard for transparency reports
├── agents/ # Agent instruction configs (YAML playbooks)
│ ├── building-inspector-monitor.yaml
│ ├── city-council-analyzer.yaml
│ ├── police-accountability.yaml
│ └── school-board-monitor.yaml
├── requirements.txt
└── setup.py
Each module is independently deployable. Start with one, add more as you go.
This project exercises existing legal rights. No new authority is needed.
| Law | Scope | What It Covers |
|---|---|---|
| Federal FOIA (5 U.S.C. § 552) | Federal agencies | Agency records, communications, internal memos |
| California Public Records Act (Gov. Code § 7920-7931) | CA state & local | All public records not specifically exempt |
| Brown Act (Gov. Code § 54950-54963) | CA local bodies | Open meetings, agendas, minutes, recordings |
| Political Reform Act (Gov. Code § 81000+) | CA officials | Form 700 financial disclosures, campaign finance |
| Body Camera Laws (varies by state) | Law enforcement | Officer-worn camera footage as public record |
| Permit Records (local ordinances) | Building departments | Permits, inspections, code enforcement actions |
The key legal concept: practical obscurity -- the idea that information is effectively private because nobody has the bandwidth to aggregate it. Courts have recognized this doctrine (DOJ v. Reporters Committee, 1989), but it was never a legal right. It was a side effect of human limitations. AI eliminates the side effect without violating any law.
# Clone
git clone https://github.com/rayhe/agent-watchdog.git
cd agent-watchdog
# Install dependencies
pip install -r requirements.txt
# Copy and customize a city config
cp config/city-template.yaml config/my-city.yaml
# Edit my-city.yaml with your city's details
# Run the FOIA engine
python -m modules.foia-engine.foia_tracker --config config/my-city.yaml
# Run the meeting monitor
python -m modules.meeting-monitor.meeting_monitor --config config/my-city.yaml
# Run the permit analyzer
python -m modules.permit-inspector.permit_analyzer --config config/my-city.yaml
# Generate the public dashboard
python -m modules.dashboard.generate --config config/my-city.yaml --output docs/Based on current API pricing (as of April 2026):
| Component | Monthly Cost (city of 80K) |
|---|---|
| LLM inference (meeting transcripts, cross-referencing) | ~$18 |
| FOIA request generation & tracking | ~$4 |
| Document OCR and extraction | ~$8 |
| Hosting (static dashboard) | ~$0 (GitHub Pages) |
| Email notifications | ~$2 (Resend free tier) |
| Storage (documents, transcripts) | ~$6 |
| Total | ~$38/month |
Scale: $740K/year covers the 20 largest California cities (180K public employees).
Generates legally compliant public records requests, tracks responses, auto-escalates when deadlines pass. Includes state-specific templates with proper statutory citations.
Ingests public meeting transcripts (city council, planning commission, school board), extracts voting records, flags unusual patterns, cross-references votes with financial disclosures.
Parses Form 700 (CA Statement of Economic Interests) and other financial disclosures. Cross-references holdings, income sources, and gifts against policy votes and contract awards.
Analyzes building permit and inspection data for anomalies: unusual approval rates by inspector, time-to-approval outliers, geographic clustering of fast-tracked permits, repeat applicant patterns.
Generates a static HTML dashboard with all findings. No editorializing -- just data, sources, and links. Designed for GitHub Pages deployment.
The agents/ directory contains YAML playbooks defining what each oversight agent monitors, what data sources it uses, and what anomalies it flags. These are designed to be human-readable and forkable.
See agents/README.md for details on writing custom agent configs.
This project needs:
- State-specific FOIA templates -- every state has different public records laws
- City configs -- the more cities covered, the better
- Data source connectors -- scrapers for specific city portals, court record systems, etc.
- Dashboard improvements -- better visualizations, mobile support
- Legal review -- if you're a lawyer, we'd love your eyes on the templates
Open an issue or submit a PR. All contributions must include source citations.
- No editorializing. The dashboard presents data and sources. It does not assign guilt or innocence.
- Always link sources. Every claim must trace back to a public document.
- Respect exemptions. When a FOIA exemption is valid (personnel records, active investigations, attorney-client privilege), respect it.
- No harassment. This tool monitors public conduct of public employees in their public roles. It does not monitor private lives.
- Corrections welcome. If data is wrong, fix it publicly and promptly.
MIT License. See LICENSE.
This project is a companion to the article Every Public Employee in America Is Already Legally Observable. AI Agents Could Actually Do It. published on Live in the Future.
The article introduces the framework of Agentic Government Oversight -- not new surveillance, but the automated exercise of transparency rights that already exist in law. The gap between what's legally public and what anyone actually monitors is enormous. This toolkit closes that gap.