Migration: Dashboard-First → Messaging-First AI Marketing Platform

Goal: Transform the current Fastify API + Vue dashboard into a messaging-first platform where GoClaw orchestrates multi-agent teams, Telegram is the primary client interface, and the Cloud Run API handles deterministic execution.

Overview

This migration is organized into 6 phases spanning 10-12 weeks. Each phase builds on the previous one, with some phases running in parallel.

Migration Phases

Current State

What exists today

What's missing

1. No Telegram integration — clients use Vue dashboard only
2. No multi-agent orchestration — single LangChain chains
3. No brand voice / long-term memory — no pgvector
4. No proactive content — no website monitoring or event calendar
5. No analytics feedback loop — data not fed back to LLM
6. No unified publish endpoint — Ghost only
7. No media ingestion — no image/document processing
8. GoClaw not connected to Supabase — standalone mode

Dependencies Graph

Phase 0 (Foundation)
  ├── 0.1 GoClaw + Supabase ─────────────────┐
  ├── 0.2 Telegram Bot ──────────────────────┐│
  └── 0.3 Config unification ────────────────┤│
                                              ││
Phase 1 (Agent Team) ◄───────────────────────┘│
  ├── 1.1 Agent specs ◄───────────────────────┘
  ├── 1.2 Skill files
  ├── 1.3 Custom tools ──────┐
  └── 1.4 Bindings           │
                              │
Phase 2 (API Expansion) ◄────┘
  ├── 2.1 New endpoints
  ├── 2.2 New DB tables
  ├── 2.3 Content refactor
  └── 2.4 Publish adapters ──────────┐
                                      │
Phase 3 (Conversation) ◄──────────────┘
  ├── 3.1 Onboarding
  ├── 3.2 Approval loop
  └── 3.3 Notifications
            │
Phase 4 (Proactive) ◄─────────────────
  ├── 4.1 Website monitoring
  ├── 4.2 Event calendar
  ├── 4.3 Analytics feedback
  └── 4.4 Scheduler overhaul

Phase 5 (Dashboard) — parallel with Phase 3-4
  ├── 5.1 Role change
  ├── 5.2 New pages
  └── 5.3 Real-time updates

Phase 6 (Production) — after all functional phases
  ├── 6.1 Multi-tenancy
  ├── 6.2 Security
  ├── 6.3 Deployment
  └── 6.4 Testing

Quick Wins (Can Do Immediately)

These tasks can be started without blocking on other work:

1. Enable pgvector in Supabase — one SQL command
2. Create telegram_client_mapping table — simple migration
3. Extract Unsplash search into its own endpoint — code exists
4. Create the GET /client/:id/context endpoint — just a query
5. Write the three SKILL.md files — no code changes
6. Set up GoClaw docker-compose overlay — config only

Dependency & Tooling Summary

New npm packages (api/)

• twitter-api-v2 — Twitter/X publishing
• cheerio + turndown — URL scraping → Markdown
• fast-xml-parser — Sitemap/RSS parsing
• @mailchimp/mailchimp_transactional — Email sending
• ioredis (optional) — Caching layer

New npm packages (front-end/)

• apexcharts + vue3-apexcharts — Analytics charts
• v-calendar — Content calendar view

Remove

• bree — Replaced by GoClaw cron system

Infrastructure

• pgvector extension in Supabase
• GoClaw managed mode
• Redis (optional) for caching
• OAuth proxy (optional) for social connections

Risk Register

Getting Started

Begin with Phase 0: Foundation & Infrastructure to wire up GoClaw with Supabase and configure the Telegram bot.