Analyse Inboxmate

InboxMate Ecosystem Reference

Use this skill to understand the InboxMate product, its architecture, outreach pipeline, and how to query data across systems.

Architecture

InboxMate is a module (chatwidget) within the AgentHub platform — a full-stack Nuxt 3 + Supabase + Socket.IO SaaS.

Repos & Key Files

URLs

Demo Pipeline

/find-leads [N]          → CRM companies + people (German B2B, UWG-compliant)
  ↓
/inboxmate-batch-demo    → Demos created, opportunities at SCREENING/PENDING_REVIEW
  ↓
/review-demos            → demoStatus: OK_TO_SEND or NEEDS_FIX
  ↓
/fix-demos               → Fixes applied, reset to PENDING_REVIEW
  ↓
/plan-campaign           → CRM campaign created, opportunities linked with shared deadline
  ↓
/setup-email-drafts      → Email drafts at notifications.psquared.dev/drafts
  ↓
Human review + send      → demoStatus: SENT, outreachSentAt timestamped
  ↓
/check-outreach-status   → Follow-up drafts created (5+ days after send)
  ↓
/check-demo-analytics    → Pipeline health report (Ackee + CRM)
  ↓
/refurbish-demos         → Upgrade knowledge for campaign demos (clear + re-scrape via MCP)

CRM State Machine (Twenty CRM)

Opportunity.demoStatus

Opportunity.demoType

Opportunity.stage

Campaign Structure

Campaigns group opportunities into sendable batches with shared offerExpiresAt and offerText. Created by /plan-campaign, consumed by /setup-email-drafts.

How to Query Each System

CRM (Twenty) — GraphQL

• Endpoint: https://crm.psquared.dev/graphql
• Auth: Authorization: Bearer $PSQUARED_CRM_TOKEN
• Key objects: companies, people, opportunities, campaigns, tasks, notes

Example — list opportunities in a campaign:

query {
  opportunities(filter: { campaignId: { eq: "CAMPAIGN_ID" } }) {
    edges { node { id name stage demoStatus demoUrl } }
  }
}

Supabase (AgentHub DB)

• Project: fevtfywriufbqnvbgyrm
• Tool: mcp__plugin_supabase_supabase__execute_sql
• Key tables: demo_pages, agents, knowledge, knowledge_bucket_items, knowledge_bucket_chunks, chats, messages, accounts

Example — find chats on demo agents:

SELECT dp.company_name, c.id as chat_id, c.created_at,
  (SELECT count(*) FROM messages m WHERE m.chat_id = c.id) as msg_count
FROM demo_pages dp
JOIN chats c ON c.agent_id = dp.agent_id
ORDER BY c.created_at DESC

Ackee Analytics — GraphQL

• Endpoint: https://ackee.psquared.dev/api
• Auth: Authorization: Bearer $ACKEE_TOKEN
• Demo domain ID: 4bdddc8c-11d9-4d7e-ab94-aeb7866f0bb2
• CTA event ID: 0f037f1c-c2c0-4205-b90d-3cb9bf66f9c2

Example — get demo page views:

query {
  domain(id: "4bdddc8c-11d9-4d7e-ab94-aeb7866f0bb2") {
    statistics { views(interval: MONTHLY, type: UNIQUE) { count } }
    facts { viewsMonth viewsToday activeVisitors }
  }
}

Notification Service — REST

• Base URL: https://notifications.psquared.dev
• Auth (admin): Authorization: Bearer $BEARER_TOKEN
• Auth (drafts only): Authorization: Bearer $EMAIL_DRAFT_ONLY_BEARER

Key endpoints:

• GET /drafts?status=SENT&pageSize=50 — list sent emails
• GET /drafts/:id — read full email (html_body, variables, subject)
• POST /drafts/create — create draft from template
• POST /drafts/send — send drafts { draftIds: [...] }
• GET /stats/funnel — full conversion funnel data
• GET /drafts/campaigns — list CRM campaigns with draft counts

InboxMate MCP — JSON-RPC

• Endpoint: https://app.psquared.dev/api/mcp
• Auth: Authorization: Bearer $NUXT_MCP_DEMO_TOKEN
• Protocol: JSON-RPC 2.0

Key tools:

• clear_bucket — remove all items from a knowledge bucket
• scrape_and_build_knowledge — scrape URLs via Tavily, create knowledge items with real sourceUrls
• create_agent, update_prompt, update_widget_style, publish_agent
• create_knowledge_bucket, add_to_bucket, set_knowledge
• create_demo_page, update_demo_page, list_demos, get_demo
• create_inbox_demo, update_inbox_demo — Demo-Postfach (type='inbox': seeded pre-triaged inbox, no agent)
• list_bucket_items, list_knowledge, get_knowledge

Key Metrics

Environment Variables

All in /Users/martinpammesberger/Documents/psquared/claude-overlord-folder/.env:

Pricing Tiers (credits model, since 2026-06)

Note: If prices change, this table must be updated. The /price-change skill handles pricing updates across all touchpoints — make sure this file is included in the update list.

Model: pay per Postfach (mailbox) + AI credits (chat reply 1, email draft 2, tool call/scrape/PDF 1; setup & rule-based sorting free). Manual invoicing — no Stripe checkout.

Top-ups: €9/€8 per 1,000 (Starter/Pro), instant grant, invoiced afterwards. All paid plans 14-day free trial, no credit card.

Knowledge System

Agents use RAG (vector search) over knowledge buckets:

• Each bucket contains knowledge items (text entries with sourceUrl)
• Items are chunked and embedded for vector similarity search
• When the agent's searchKnowledge tool is called, it returns matching chunks WITH their sourceUrl
• The widget UI shows sources as clickable links in citations

To upgrade knowledge quality: clear_bucket → scrape_and_build_knowledge with real page URLs → publish_agent