Skip to content

LAUNCH_PLAN.md

Latest commit

 

History

History
245 lines (188 loc) · 8.08 KB

LAUNCH_PLAN.md

File metadata and controls

245 lines (188 loc) · 8.08 KB

Launch Plan: BillionMail + FrostByte Go-Live

Summary

FrostByte at /Users/imorgado/Desktop/Workspace/projects/frostbyte/ is production-ready (14 integrations). Needs: LLM layer refactor for GLM-4.7, API key setup via Chrome DevTools, VPS provisioning, domain/DNS creation, frontend fixes.

Monthly Cost: ~$56/mo at launch, ~$95/mo with video

Service Cost Notes
Hetzner VPS (CX22) $7-12 2 vCPU, 4GB RAM, Ubuntu
GLM-4.7 $0 Yearly plan, OpenAI-compatible API
Apollo.io Basic $49 5K credits + unlimited emails
Cal.com (free cloud) $0 30-min meetings, Google Calendar
RapidAPI LinkedIn $0 Free tier
Cartesia (voice) Usage-based TTS + voice clone
SyncLabs (lip sync) Usage-based Lip sync API
Cloudflare R2 ~$0-5 Video/thumbnail storage
Launch total ~$56-61
With video ~$65-80 Depends on volume

PHASE 1: Refactor FrostByte LLM Layer [DONE]

Completed in dc2f622. Provider abstraction added, GLM-4.7 support via OpenAI-compatible client. 351 tests pass.

Changes made:

  • frostbyte/personalization/llm.py — LLMProvider ABC + AnthropicProvider, GroqProvider, OpenAICompatProvider. Config-driven routing with fallback.
  • frostbyte/config.py — Added glm_api_key, glm_base_url, glm_model, first_touch_provider, followup_provider, classify_provider
  • pyproject.toml — Added openai>=1.0.0
  • .env.example — Added GLM env vars
  • Tests updated to match new provider architecture

PHASE 2: Link FrostByte + VPS Setup

2a. Symlink FrostByte

ln -s /Users/imorgado/Desktop/Workspace/projects/frostbyte \
      /Users/imorgado/Desktop/Workspace/projects/BillionMail/frostbyte

Verify Dockerfile EXPOSE vs docker-compose port 8001.

2b. Provision Hetzner VPS (Chrome DevTools)

  1. Navigate to hetzner.com -> Cloud Console
  2. Create CX22 server (2 vCPU, 4GB, Ubuntu 22.04)
  3. Note public IP
  4. Configure rDNS to match sending domain

2c. Create Josh's Workspace Domain (Chrome DevTools)

  1. Navigate to admin.google.com -> Account -> Domains -> Manage domains -> Add domain
  2. Purchase/add sending domain
  3. Verify domain ownership (DNS TXT record)
  4. Create josh@ user account
  5. This is a prerequisite for DNS records + Gmail app password

2d. DNS Records for Sending Domain

On the domain registrar, add:

  • SPF: v=spf1 ip4:<vps-ip> include:_spf.google.com ~all
  • DKIM: Generated by BillionMail after deploy
  • DMARC: v=DMARC1; p=none; rua=mailto:dmarc@<domain>
  • MX: Point to VPS IP
  • rDNS: Set on Hetzner to match domain

PHASE 3: API Keys via Chrome DevTools

Use Chrome DevTools MCP to log into each dashboard and get keys:

# Service URL Free?
1 GLM / Z.AI api.z.ai Yes (yearly plan)
2 Apollo.io app.apollo.io -> Settings -> API $49/mo Basic
3 RapidAPI rapidapi.com -> Fresh LinkedIn Profile Data Free tier
4 Cal.com cal.com -> Settings -> Security -> API Key Free
5 Gmail App Password myaccount.google.com -> Security -> App Passwords Free
6 Cartesia cartesia.ai (voice clone + TTS) Usage-based
7 SyncLabs synclabs.so (lip sync) Usage-based
8 Cloudflare R2 dash.cloudflare.com (video storage) ~$0-5/mo

Write to /Users/imorgado/Desktop/Workspace/projects/frostbyte/.env:

# LLM (GLM-4.7 via yearly plan)
GLM_API_KEY=<from zhipu>
GLM_BASE_URL=https://api.z.ai/api/paas/v4/
GLM_MODEL=glm-4.7
FIRST_TOUCH_PROVIDER=glm
FOLLOWUP_PROVIDER=glm
CLASSIFY_PROVIDER=glm

# Keep as fallbacks (empty until needed)
ANTHROPIC_API_KEY=
GROQ_API_KEY=
OPENAI_API_KEY=

# Enrichment
APOLLO_API_KEY=<from apollo>
RAPIDAPI_KEY=<from rapidapi>
HUNTER_API_KEY=
PROSPEO_API_KEY=

# Video Pipeline (Cartesia + SyncLabs + R2)
ANTHROPIC_API_KEY=
CARTESIA_API_KEY=
SYNCLABS_API_KEY=
R2_ACCOUNT_ID=
R2_ACCESS_KEY_ID=
R2_ACCESS_KEY_SECRET=
R2_BUCKET_NAME=
R2_PUBLIC_URL=
VOICE_SAMPLE_URL=
VOICE_DEFAULT_ID=

# Booking
CALCOM_API_KEY=<from cal.com>

# Gmail
GMAIL_APP_PASSWORDS={"josh@<domain>":"<app password>"}

# Config
SENDING_BACKEND=billionmail
SECRET_KEY=<generate>
NOTIFICATION_EMAIL=josh@<domain>
SENDER_NAME=Josh Izzy
SENDER_COMPANY=Douro Digital

PHASE 4: Configure BillionMail UI (Chrome DevTools)

  1. Navigate to BillionMail instance on VPS
  2. Login
  3. Settings > AI Model: Add GLM-4.7 supplier (base URL + API key)
  4. Settings > SMTP Relay: Configure Gmail relay
  5. Domain Management: Add sending domain, generate DKIM keys, verify DNS

PHASE 5: Frontend Fixes (Code Changes)

5a. Remove automation stub

  • Delete: core/frontend/src/views/automation/index.vue
  • Delete: core/frontend/src/router/modules/automation.ts
  • Remove import from router index

5b. Enable sequences create

  • File: core/frontend/src/views/sequences/list/index.vue:4 — remove disabled
  • File: core/frontend/src/api/modules/sequences.ts — add createSequence(), toggleSequence()
  • New: core/frontend/src/views/sequences/list/components/SequenceForm.vue (modal pattern)

5c. Cleanup

  • Delete: core/frontend/src/views/contacts/group/temp.vue

5d. i18n

  • Files: core/frontend/src/i18n/lang/{en,zh,ja}.json — sequence form keys

PHASE 6: Cal.com + Google Calendar (Chrome DevTools)

  1. Josh creates Cal.com account with Workspace email
  2. Connect Google Calendar via OAuth
  3. Create "30-Minute Meeting" event type
  4. Copy API key + event type ID -> frostbyte .env
  5. Apple Calendar: syncs via Google Calendar bridge

PHASE 7: Deploy & Verify

  1. Push code to VPS, docker-compose up -d
  2. Verify FrostByte: curl http://<vps>:8001/health
  3. Test sequence creation from frontend
  4. Test lead scraping via enrichment UI
  5. Send test email through sequence engine
  6. Verify tracking webhooks

Verification Checklist

  • VPS provisioned, rDNS configured
  • Domain created, DNS records set (SPF/DKIM/DMARC)
  • FrostByte containers running (api, worker, beat)
  • /health returns 200
  • GLM-4.7 responds via LLM layer
  • Sequences list loads + create works
  • Leads list loads
  • Enrichment scraper runs
  • Gmail relay sends test email
  • Cal.com booking link -> Google Calendar
  • Automation route gone from nav

POST-LAUNCH BACKLOG

Blocked on Josh

  • Record voice sample (clear audio, 30s+) for Cartesia voice clone
  • Set VOICE_SAMPLE_URL or VOICE_DEFAULT_ID in env
  • Create Cloudflare R2 bucket + generate API tokens

How Video Personalization Works (Built in Go)

Full pipeline implemented in core/internal/service/video_gen/. Per contact:

  1. Enrichment populates lead data (website_url, business_name, signals)
  2. Screenshots — Playwright captures homepage, contact, Google pages
  3. Annotate — ImageMagick overlays signal callouts on screenshots
  4. Script — Claude Sonnet generates personalized 15-30s narration
  5. Voice — Cartesia TTS with cloned or default voice
  6. Composite — FFmpeg stitches annotated images + audio into video
  7. Lip Sync — SyncLabs adds lip sync (optional, if API key set)
  8. Upload — Cloudflare R2 storage, generates landing page URL
  9. Update — Stores video_url, thumbnail_url, landing_page_url in contact attribs

Feature Gaps

  • Leads detail/edit view (frontend only has list)
  • Sequences edit UI (can create but not edit)
  • Enrichment data results view (UI shows config only)
  • Richer video scripts using full enrichment data

Bug Fixes

  • Excel export returns 500 — core/internal/controller/contact/contact_v1_export_contacts.go:86
  • SSL cert listing returns CodeNotImplemented

Future Upgrades

  • Anthropic Claude as fallback LLM (if GLM quality insufficient for English copy)
  • Apollo upgrade tiers as credit usage grows
  • Alternative TTS providers as fallback to Cartesia

Resolved

  • GLM endpoint: api.z.ai (international), model: glm-4.7
  • Domain: needs creation in Google Workspace (done in Phase 2c via Chrome DevTools)

Unresolved

  • What domain name for Josh's sending email? (need to pick/purchase before Phase 2c)