Instant Nonprofits — AI Agent Architecture Proposal
Purpose: Define what agent system we would propose to Instant Nonprofits, which platforms to use, and how to phase the buildout. Prepared: 2026-03-16 For internal use: Solanasis team prep — DO NOT share directly with client. Inform the proposal, not replace it. Related: Email & CRM Dashboard Solutions | Discovery Call Prep
What Problem Are We Solving with Agents?
Christian’s team of 5-6 people needs to operate like a team of 20. Specifically:
- Email Triage Agent — Classify, prioritize, and route inbound emails so nothing gets missed
- Follow-Up Agent — Monitor for stale conversations and auto-trigger or auto-draft follow-ups
- Client Intake Agent — Run onboarding questionnaires, parse responses, populate systems
- Data Enrichment Agent — Process the 5,000-client database, find signals, score for upsells
- Upsell Detection Agent — Identify which existing clients need additional services and why
- Personal Assistant Agent — Christian’s “someone checking my emails” — a daily briefing, priority list, and action items
These aren’t separate products. They’re workflows that use AI reasoning to make decisions a human would normally make, running on automation platforms.
Platform Decision: n8n vs. LangGraph vs. Make.com vs. Others
The Short Answer
For Instant Nonprofits today: Make.com AI Agents + n8n for complex workflows. For Solanasis to build as a repeatable service: n8n (self-hosted) + Claude Agent SDK.
Here’s why.
Platform Comparison for This Use Case
| Criteria | Make.com | n8n (Cloud) | n8n (Self-Hosted) | LangGraph | CrewAI |
|---|---|---|---|---|---|
| Technical complexity | Low | Medium | Medium-High | High | Medium-High |
| Can the client maintain it? | Yes | Mostly | No | No | No |
| Gmail integration | Native | Native | Native | Custom code | Custom code |
| ClickUp integration | Native | Native | Native | Custom code | Custom code |
| Slack integration | Native | Native | Native | Custom code | Custom code |
| Clay.com integration | HTTP API | HTTP API | HTTP API | Custom code | Custom code |
| WordPress integration | Native | Native | Native | Custom code | Custom code |
| AI agent capabilities | Good (Oct 2025 release) | Excellent | Excellent | Best-in-class | Excellent |
| Multi-step workflows | Good | Excellent | Excellent | Best-in-class | Good |
| Monthly cost (platform) | 29 + credits | 50 | Free (infra: 500) | Free (infra: 500) | Free (infra: 500) |
| Self-hosted option | No | No | Yes | Yes | Yes |
| Visual builder | Yes | Yes | Yes | No | No |
| Solanasis can white-label | No | Possible | Yes | Yes | Yes |
Why NOT LangGraph or CrewAI Right Now
LangGraph is the gold standard for complex multi-agent orchestration. It models agent workflows as directed graphs with persistent state, checkpointing, and branching. But:
- Requires solid Python development skills
- Steep learning curve (graph theory, state machines)
- Christian’s team can’t maintain it
- Overkill for the workflows we’re building initially
- Best suited for: complex reasoning chains, multi-step research agents, autonomous planning
CrewAI is simpler than LangGraph and uses a role-based metaphor (assigning agents “roles” and “goals”), but still:
- Requires Python
- Less mature in production
- Client can’t maintain it
When to revisit these: Once Instant Nonprofits has validated their workflows on Make.com/n8n and we have clear evidence that the visual builders can’t handle the complexity they need (likely 6-12 months in).
Why Make.com + n8n is the Right Call
Make.com handles 80% of what they need:
- Email → classify → route (native Gmail + AI Agent nodes)
- Form submission → ClickUp task creation (native integrations both sides)
- Stale task → Slack notification (ClickUp → delay → Slack)
- Meeting recorded → follow-up drafted (Fathom webhook → AI → Gmail)
- Contact enriched → update CRM (Clay API → ClickUp/HubSpot)
n8n handles the remaining 20% where Make.com falls short:
- Complex multi-step AI reasoning (data enrichment with multiple API calls)
- Self-hosted option for cost control at scale (important with 5,000 clients)
- More flexible AI agent nodes (supports Claude, GPT, local models)
- Better for batch processing (running enrichment on 5,000 records)
- White-label potential (Solanasis can host n8n for clients as a managed service)
The hybrid approach:
- Quick automations (email routing, Slack alerts, form handling) → Make.com
- AI-heavy workflows (enrichment, scoring, intelligent triage) → n8n
- Both connect to the same tools (ClickUp, Gmail, Slack, Clay)
- Christian’s team manages Make.com; Solanasis manages n8n
The Agent System: What We’d Build
Agent 1: Email Triage & Priority Agent
Platform: Make.com AI Agent → Phase 2: n8n with Claude
What it does:
- Monitors Christian’s Gmail inbox (and Jackie’s, if needed)
- AI classifies each inbound email into categories:
- HOT — Investor, major partner, time-sensitive deal (notify Slack immediately)
- WARM — Existing client needing help, referral intro, follow-up on active deal
- ROUTINE — Newsletter, vendor pitch, informational, can-wait
- NOISE — Spam, cold outreach, irrelevant
- Creates a daily priority digest (Slack message or ClickUp dashboard)
- Auto-drafts responses for ROUTINE emails (human reviews before sending)
- Auto-archives NOISE
Example flow:
Email arrives → AI reads subject + body + sender history
→ Is sender in ClickUp/CRM? → If yes, pull context
→ Classify priority → Route:
HOT → Slack #urgent + ClickUp task + auto-draft response
WARM → Slack #inbox + ClickUp task
ROUTINE → Daily digest + auto-draft response (hold for review)
NOISE → Archive + unsubscribe suggestion
Expected impact: Christian goes from “300 unread emails, who knows what’s important” to “here are your 5 urgent items for today with draft responses.”
Build time: 1-2 weeks Monthly cost: 30 (Make.com) + 20 (AI API calls)
Agent 2: Follow-Up Guardian Agent
Platform: Make.com → Phase 2: n8n
What it does:
- Monitors all outbound emails for replies
- If no reply received within X days (configurable per contact type), triggers follow-up
- Auto-drafts a contextual follow-up email (references original message)
- Escalates to Slack if follow-up goes unanswered after 2nd touch
- Logs all follow-up activity in ClickUp
Example flow:
Outbound email sent → Start timer (3 days for investors, 5 days for clients, 7 days for cold)
→ No reply received? → Auto-draft follow-up
→ Send to Slack for Christian to review + approve
→ Christian approves → Send follow-up
→ Still no reply after 2nd follow-up → Create ClickUp task: "Decide: pursue or close"
Christian-specific use case: He told David Meltzer’s team he’d send the investor deck and took 2 weeks. This agent would have flagged it on Day 3.
Build time: 1 week Monthly cost: 15 (Make.com) + minimal AI cost
Agent 3: Client Intake & Onboarding Agent
Platform: Make.com + WordPress form (or Typeform/Tally)
What it does:
- New client fills out intake form on website
- AI parses responses and extracts: entity type, state, budget, urgency, special needs
- Creates structured ClickUp task with all parsed fields
- Assigns to appropriate team member based on entity type
- Sends personalized welcome email sequence (3-5 emails over 2 weeks)
- Schedules kickoff call automatically
Example flow:
Client fills form → AI extracts data + classifies service tier
→ Create ClickUp task (assigned to Catherine or Jackie based on type)
→ Send welcome email #1 (immediate)
→ Schedule email #2 (Day 3: "Here's what to expect")
→ Schedule email #3 (Day 7: "Your formation checklist")
→ If nonprofit formation: auto-generate document checklist
→ If compliance renewal: auto-pull existing record from database
Key discovery question: What questions SHOULD they be asking in the intake form that they’re not asking now? This is where we add value by designing a smarter questionnaire that:
- Pre-qualifies for service tier
- Identifies upsell opportunities at intake (before the engagement even starts)
- Collects data needed for formation (reducing back-and-forth)
- Gauges tech needs (do they need a website? Email? Donor management?)
Build time: 2-3 weeks Monthly cost: 30 (Make.com + form tool)
Agent 4: Database Enrichment Agent
Platform: n8n + Clay.com API + Claude Agent SDK
What it does:
- Takes the 5,000-client database and processes it in batches
- For each nonprofit, queries: IRS TEOS (Tax Exempt Organization Search) status, website status, social media, GuideStar/Candid profile, recent 990 filings, board members, annual revenue
- Classifies each nonprofit into segments:
- Thriving — Active, growing, likely has budget for additional services
- Stable — Active, maintaining, may need compliance help
- Struggling — Active but problems (website down, no recent filings, leadership changes)
- Dormant — No recent activity, may have dissolved
- Unknown — Insufficient data to classify
- Scores each for upsell readiness (1-10)
- Outputs enriched data back into ClickUp or CRM
This is the big-ticket project. The 5,000-client database is Instant Nonprofits’ most valuable asset, and it’s currently sitting idle.
Data sources for enrichment:
- IRS TEOS/BMF — Is the nonprofit still tax-exempt? When was their last filing?
- GuideStar/Candid — Revenue, programs, leadership
- Clay.com enrichment — Contact info, social profiles, company data
- Website check — Is their website up? When was it last updated? Does it have SSL?
- Social media scan — Do they have LinkedIn, Facebook, Instagram? Are they active?
- State registry — Are they in good standing with their state?
Build time: 3-4 weeks (including data modeling and testing) Monthly cost: 500 (Clay.com credits + AI API calls + n8n infrastructure)
Revenue potential: If enrichment reveals that even 500 of the 5,000 (10%) are thriving nonprofits that need additional services at 5,000 each, that’s 2.5M in pipeline from an asset that’s currently generating $0.
Agent 5: Upsell Detection Agent
Platform: n8n + Claude
What it does:
- Monitors enriched client data for trigger events:
- Website goes down or has security issues → “Offer website management package”
- Board member changes → “Offer governance review + compliance check”
- Annual filing deadline approaching → “Offer annual compliance package”
- No 990 filed in 2+ years → “Flag: may have compliance problems, outreach needed”
- Revenue growth detected → “Offer strategic advisory or tech setup”
- New social media activity → “Offer digital marketing consultation”
- Creates opportunity records in CRM/ClickUp
- Auto-drafts personalized outreach based on the specific trigger
- Routes to salesperson (Jackie or Christian) with full context
Example output:
🔔 UPSELL ALERT: Hope Foundation (formed 2021)
Trigger: Website SSL certificate expired + no 990 filed in 2024
Suggested action: Compliance check package ($1,500) + website fix ($500)
Draft email: "Hi [Name], we noticed your website might have a security issue
and wanted to check in on your annual compliance. We can help with both..."
→ Assigned to: Jackie | Priority: High | ClickUp task created
Build time: 2 weeks (after enrichment agent is running) Monthly cost: 50 (minimal — mostly AI reasoning on existing data)
Agent 6: Personal Briefing Agent (Christian’s “Email Butler”)
Platform: n8n or Make.com → Slack or email
What it does:
- Runs every morning at 7 AM
- Scans: Gmail (new + awaiting reply), ClickUp (overdue tasks, today’s tasks), Calendar (today’s meetings)
- Generates a personalized daily briefing:
- Top 3 Priority Items (based on urgency, revenue impact, relationship importance)
- Emails Needing Response (ranked, with draft responses)
- Today’s Meetings (with 1-liner prep notes)
- Follow-Ups Overdue (who you owe something to)
- Quick Wins (tasks you can close in under 5 minutes)
- Delivers via Slack DM or dedicated Slack channel
Example daily briefing:
☀️ Good morning, Christian — Monday March 16
🔴 URGENT (respond today):
1. David Meltzer's team — waiting on your investor deck (overdue 12 days)
→ Draft response attached. Takes 2 min to send.
2. New lead from Aspire Tour — interested in fiscal sponsorship
→ Jackie has context. Schedule 15-min call? [Calendly link]
3. Catherine flagged: WordPress site throwing 500 errors intermittently
→ Dmitri (Solanasis) is on it.
📧 INBOX SUMMARY: 47 new emails | 5 need response | 12 FYI | 30 noise (auto-archived)
📅 TODAY'S CALENDAR:
- 10:00 AM: Jackie 1:1 (ClickUp review + new hires)
- 2:00 PM: Solanasis discovery call (Dmitri)
- 4:30 PM: Networking happy hour (no prep needed)
⏰ OVERDUE FOLLOW-UPS:
- Marcus Lemonis intro (11 days) — draft follow-up ready
- GoFundYourself production team (8 days) — draft follow-up ready
- Website redesign vendor (5 days) — draft follow-up ready
✅ QUICK WINS (under 5 min each):
- Approve Catherine's weekly report
- Respond to Slack DM from new marketer
- Review and sign e-doc from accountant
This is the “killer feature.” When Christian opens his laptop, instead of facing 300 unread emails and anxiety, he gets a crisp, prioritized action list. This is what he meant by “I need someone checking my emails.”
Build time: 1-2 weeks Monthly cost: 30 (AI API calls + automation platform)
Phased Buildout Plan
Phase 1: Quick Wins (Weeks 1-3) — 1,500 to build
| Agent | Platform | Impact | Build Time |
|---|---|---|---|
| Personal Briefing Agent (#6) | Make.com | Immediate relief for Christian | 1 week |
| Follow-Up Guardian (#2) | Make.com | Stops deals from dying | 1 week |
| Basic email triage (#1 lite) | SaneBox + Make.com | Inbox under control | 3 days |
Why start here: These deliver immediate, visible value. Christian will feel the difference on Day 1. This is how we earn the right to build Phase 2.
Phase 2: Revenue Engine (Weeks 4-8) — 5,000 to build
| Agent | Platform | Impact | Build Time |
|---|---|---|---|
| Client Intake Agent (#3) | Make.com + form | Automated onboarding | 2-3 weeks |
| Full Email Triage Agent (#1) | n8n + Claude | AI-powered inbox management | 2 weeks |
Why Phase 2: These require more design work (intake questionnaire design, classification logic) and benefit from what we learn in Phase 1.
Phase 3: The Gold Mine (Weeks 9-16) — 10,000 to build
| Agent | Platform | Impact | Build Time |
|---|---|---|---|
| Database Enrichment Agent (#4) | n8n + Clay.com + Claude | 5,000-client database activated | 3-4 weeks |
| Upsell Detection Agent (#5) | n8n + Claude | Revenue pipeline from existing clients | 2 weeks |
Why Phase 3: This is the highest-value but most complex work. It requires a clean CRM structure (from Phase 1-2), enrichment data (Clay.com), and validated classification logic. But the ROI is enormous — potentially $500K+ in pipeline.
Cost Summary
Build Costs (One-Time)
| Phase | Scope | Estimated Cost |
|---|---|---|
| Phase 1 | Quick wins: briefing, follow-up, basic triage | 1,500 |
| Phase 2 | Intake automation, full email triage | 5,000 |
| Phase 3 | Database enrichment, upsell detection | 10,000 |
| Total | Full 6-agent system | 16,500 |
Ongoing Monthly Costs
| Item | Monthly Cost |
|---|---|
| Make.com (Teams) | $29 |
| n8n Cloud (or self-hosted infra) | 500 |
| AI API calls (Claude/GPT) | 200 |
| Clay.com (enrichment credits) | Existing subscription + overages |
| SaneBox or Superhuman | 30 |
| Total ongoing | 759/mo |
ROI Projection
| Investment | Potential Return |
|---|---|
| 16,500 build + ~$300/mo ongoing | 5,000 clients × 10% activated × 500,000 in pipeline** |
| Break-even: 1-2 upsell closes | Long-term: 250K/year in reactivated client revenue |
What This Means for Solanasis
Why We Build This as a Service Offering
This exact agent architecture — email triage, follow-up automation, CRM integration, database enrichment, upsell detection — is what every small business with a client list needs. Instant Nonprofits becomes our proof of concept.
The play:
- Build it for Instant Nonprofits (get paid to learn)
- Document every workflow, every decision, every config as an SOP
- Package it as a “Client Reactivation Engine” or “Operations Automation Package”
- Offer it to every other Solanasis client — wealth managers, MSPs, professional services firms
- Eventually: productize it as a managed service (“We’ll run your AI operations for $X/month”)
What We Need to Know Before Building
These questions should be validated in the discovery call:
- What email provider are they on? (Gmail / Google Workspace / Outlook / Exchange)
- How many email accounts need triage? (Just Christian? Jackie too? A shared inbox?)
- What’s the actual volume? (Emails per day/week across the team)
- What ClickUp plan are they on? (Free vs. paid matters for API access and automation limits)
- What Clay.com plan are they on? (Credit limits affect enrichment capacity)
- Do they have any existing automations running? (Zapier, Make, anything?)
- What’s the database actually stored in? (ClickUp? Google Sheets? A CSV somewhere? An actual database?)
- Who would be the “agent manager” on their team? (Someone needs to review auto-drafted emails and approve follow-ups)
n8n vs. LangGraph: The Deeper Decision
When n8n is Right (Now)
n8n is the right choice today because:
- Visual workflow builder — Catherine or Jackie can see what’s happening, even if they can’t modify it
- 500+ native integrations — no custom code needed for Gmail, ClickUp, Slack, WordPress
- AI agent nodes — supports Claude, GPT, and local models natively
- Self-hosted option — run on a $20/mo VPS (Virtual Private Server) with unlimited executions
- Community templates — hundreds of pre-built agent workflows to start from
- Solanasis can manage it — we host and maintain it as part of the fractional CIO retainer
When LangGraph Becomes Right (Later)
LangGraph would make sense when:
- The agent workflows become too complex for visual builders (10+ decision branches, multi-step reasoning with memory)
- We need autonomous planning — agents that decide their own next steps rather than following predefined paths
- We’re building the Iconic Impact fiscal sponsorship platform — which will need custom agent logic for donor matching, compliance checking, and fund disbursement
- We’ve hired a developer (or Solanasis has a contractor) who can maintain Python code
LangGraph strengths that matter for the future:
- Persistent checkpointing (pause/resume long-running agent tasks)
- State serialization (audit trail of every decision)
- Graph-based control flow (if/else/loop at the agent level)
- Human-in-the-loop at any node (agent pauses for human approval, then continues)
The Bridge: n8n → LangGraph
n8n actually supports calling LangGraph agents via HTTP. So the migration path is:
- Now: Build everything in n8n with simple AI agent nodes
- Later: Replace specific n8n nodes with LangGraph agent calls when complexity demands it
- Eventually: Move the most complex workflows entirely to LangGraph, with n8n handling orchestration and integration
This is a clean, low-risk evolution path. No rip-and-replace needed.
Architecture Diagram (Conceptual)
┌─────────────────────────────────────────────────────────┐
│ INSTANT NONPROFITS │
│ Agent System │
├─────────────────────────────────────────────────────────┤
│ │
│ ┌──────────┐ ┌──────────────┐ ┌──────────────┐ │
│ │ Gmail │───▶│ Make.com │───▶│ ClickUp │ │
│ │ Inbox │ │ (Router) │ │ (CRM/PM) │ │
│ └──────────┘ └──────┬───────┘ └──────────────┘ │
│ │ │
│ ▼ │
│ ┌──────────────┐ │
│ │ n8n │ │
│ │ (AI Agents) │ │
│ └──────┬───────┘ │
│ │ │
│ ┌─────────────┼─────────────┐ │
│ ▼ ▼ ▼ │
│ ┌────────────┐ ┌──────────┐ ┌──────────────┐ │
│ │ Claude API │ │ Clay.com │ │ Slack │ │
│ │ (Reasoning) │ │ (Enrich) │ │ (Alerts) │ │
│ └────────────┘ └──────────┘ └──────────────┘ │
│ │
│ ┌───────────────────────────────────────────────────┐ │
│ │ DAILY BRIEFING OUTPUT │ │
│ │ ☀️ Priority Items | 📧 Inbox Summary | │ │
│ │ 📅 Calendar | ⏰ Overdue | ✅ Quick Wins │ │
│ └───────────────────────────────────────────────────┘ │
│ │
└─────────────────────────────────────────────────────────┘
Final Recommendation
Start with Make.com AI Agents for quick wins, add n8n for AI-heavy workflows, plan for LangGraph down the road.
This gives Instant Nonprofits:
- Immediate relief (Week 1)
- Revenue activation (Month 2-3)
- Long-term competitive advantage (Year 1+)
And it gives Solanasis:
- A proof of concept for our agent-building service
- A repeatable playbook for future clients
- A case study with real ROI numbers
- Monthly recurring revenue from hosting and managing the agent system
Pro Tip: The biggest mistake with agent systems is building too much before validating. Build Agent #6 (Personal Briefing) first — it’s the simplest, most visible, and most emotionally impactful. When Christian opens his laptop to a clean priority list instead of 300 unread emails, he’ll be sold on the rest. Let that first win fund and justify everything that follows.