How do I install CivicRecords AI?

[scottconverse]

Q: What do I need to get CivicRecords AI running on my city's server?

A: You need a machine with Docker installed. Recommended spec: 8+ cores, 32 GB+ RAM (hard minimum per the locked Windows target profile; 64 GB recommended), 50 GB+ free disk space. Target OS for the Windows installer is Windows 11 Pro 23H2+; CPU-only is supported (GPU optional).

Two install paths as of v1.2.0:

1. Windows — double-click .exe installer (new in v1.2.0).

Download the real installer from the v1.2.0 release page:

• CivicRecordsAI-1.2.0-Setup.exe
• CivicRecordsAI-1.2.0-Setup.exe.sha256 (verify with Get-FileHash .\CivicRecordsAI-1.2.0-Setup.exe -Algorithm SHA256 and compare)

Unsigned by design — Windows SmartScreen will show "Windows protected your PC — Unknown publisher" on first run. That's expected. Remediation: click More info → Run anyway, then confirm UAC. Full walkthrough in installer/windows/README.md.

After install you get two Start Menu shortcuts:

• Start CivicRecords AI — daily use. Runs docker compose up -d, opens the admin panel at http://localhost:8080. Does NOT re-run the installer, NOT re-pull any model, NOT re-seed.
• Install or Repair CivicRecords AI — full bootstrap. Prereq check → model picker → ollama pull of the selected model + nomic-embed-text embedder → baseline seed → browser.

2. Linux / macOS — guided script (unchanged).

Clone the repo at the v1.2.0 tag and run ./install.sh. Native .deb / .pkg / .dmg parity is explicit follow-on work, not shipped in v1.2.0.

Both paths handle:

• Docker detection (installs if needed on Linux; detects Docker Desktop on Windows and prints remediation if missing — does NOT silently install Docker Desktop)
• Hardware probe (CPU, RAM, optional AMD/NVIDIA GPU)
• GPU acceleration when available (ROCm on Linux, DirectML via host Ollama on Windows; Ollama native CUDA on NVIDIA)
• All 7 Docker services (postgres, redis, ollama, api, worker, beat, frontend)
• Database migrations and first-boot baseline seed (175 state exemption rules + 5 compliance templates + 12 notification templates)
• Admin account creation with a strong generated password
• New in v1.2.0: Fernet ENCRYPTION_KEY auto-generated on fresh install for at-rest encryption of connector credentials. Back this up separately from your database — losing it means losing credential access.
• New in v1.2.0: 4-model Gemma 4 picker (default gemma4:e4b; gemma4:e2b / gemma4:26b / gemma4:31b also selectable)

After install, open http://localhost:8080 and sign in with the admin credentials the installer printed. Full details in the README and USER-MANUAL.