Open patterns, starter repos, and a guided setup skill for co-building practical AI assistants with nonprofits — from a one-room workshop to a system the organization owns, runs, and maintains for itself.
Maintained by npaiadvisor (Brian Flett, in partnership with Management Advisory Services). Built in the open — contributions from fellow nonprofit-AI practitioners are welcome.
Status: early. The repos are being populated from two production systems. Watch this space or open an issue to get involved.
Most "AI for nonprofits" pilots die at handoff: the consultant leaves and the system rots because no one on staff can change it. This commons is built around the opposite goal — self-sufficiency as the deliverable. Every pattern here is designed so a non-technical nonprofit can operate, maintain, and evolve its assistant using a coding agent it talks to in plain English, with the safety rails living in the plumbing (preview-before-production), not in trusting any one AI.
Real nonprofit assistants fall into one of two shapes. The commons ships a starter for each.
| Shape A — Custom app | Shape B — Knowledge-base assistant | |
|---|---|---|
| Stack | Next.js + Vercel + Neon + OpenRouter | A hosted assistant (Claude Project / ChatGPT / Copilot / Gemini) + a knowledge base |
| UI | An admin dashboard | Chat is the UI — nothing to host |
| Best for | Monitor sources → AI brief → human-reviewed action | Q&A grounded in the org's own systems + guided, confirmation-gated actions |
| Rough cost | ~$40–100/mo | ~$8/seat, no hosting |
| Starter | app-starter |
kb-assistant-starter |
discovery(this repo) — the guided setup skill + the methodology library.app-starter— Shape A: a custom AI web-app, de-domained from a production donor-outreach system.kb-assistant-starter— Shape B: an LLM-agnostic knowledge-base assistant.
A Claude Code skill (in .claude/skills/np-ai-discovery/) that walks a nonprofit from idea to scaffolded assistant:
- Scope the first useful application (map the work → find the time-sink → name the objective).
- Interview the environment — office suite (Microsoft 365 / Google / neither), CRM, the LLM you'll use, email sender.
- Pick the shape (A or B) from the scoped problem.
- Scaffold the right starter — write the config, select the connectors, and lay out the handoff runbook.
Status: in development. See docs/ for the methodology it operationalizes.
The docs/ folder holds the sanitized, reusable version of the method behind the starters — the adoption journey, the productionalization model, and the platform/subscription decision guide. (Population in progress.)
See CONTRIBUTING.md. Anyone may open issues and PRs. Everything here must pass SANITIZATION.md — no client data, no secrets.
Code is MIT. Documentation and case studies are CC BY-SA 4.0. Authors retain copyright; the license is a grant, not a transfer.