Section 01

THE BRAND

Forward Assist isn't just a ticket app. It's a ticket app that knows what a forward assist actually is. Mil-spec aesthetic. Meme-lord energy. Professional where it counts, irreverent everywhere else.

Brand Identity

Name references AR-15/M16 forward assist button — "tap it when things won't go into battery"
Dark, tactical aesthetic — think armory meets SaaS dashboard
Stencil typography (Bebas Neue) for headers — feels stamped on a pelican case
Monospace (JetBrains Mono) for data — like reading a mil-spec data sheet
Humor is baked into the UI language, not just slapped on

Tone & Voice

Self-aware and irreverent — Gun Slaps "your wife's boyfriend" energy
Inside jokes the community gets: "send it", "LARP", "just as good", "two world wars"
Professional in customer-facing responses — the memes stay internal
Never punching down — humor is for the team, not at customers' expense
Think: if Garand Thumb built a SaaS product

Target Audience

Small-to-mid firearms manufacturers and parts makers
Mil-spec / defense contractors with small sales teams
Gun shop owners who are terminally online
Anyone in the industry who's tired of Zendesk looking like a dentist's office

Competitive Edge

No other ticket system speaks this language
AI that actually understands firearms context (parts, specs, compliance)
Branding that makes the team WANT to use the tool
Self-hosted option — these guys don't love cloud dependencies

Color Palette — "Armory After Dark"

OD Green + FDE + midnight. If Magpul made a dashboard.

Midnight

#0D0D0D

Cerakote Dark

#1E1E1E

OD Green

#4A5D23

OD Light

#6B8C3E

FDE

#C4A265

Coyote

#8B7355

Tracer Red

#C0392B

Stainless

#E8E8E8

UI Language — "Boring to Based"

Every button, label, and status is an opportunity.

Submit → SEND IT

Approved → CLEARED HOT

Rejected → MISFIRE

Pending Review → IN THE CHAMBER

Closed → BRASS POLICED

Urgent → DANGER CLOSE

New Ticket → INCOMING

Assigned → ON TARGET

Dashboard → THE ARMORY

Settings → FIELD STRIP

AI Draft → SPOTTER CALL

Knowledge Base → THE MANUAL

Tagline Candidates

"When your tickets won't chamber."

Primary — references the literal forward assist function

"The help desk that slaps."

Secondary — double meaning with Gun Slaps reference

"Customer service. Full send."

For marketing — "send it" is universal gun meme

"Smack that bolt home."

Aggressive variant — for stickers/merch

"Your AI spotter never misses."

AI-focused — for the tech-forward pitch

"Steal underpants → Forward Assist → Profit"

Internal only — Eric knows what's up

Section 02

THE WORKFLOW

Email comes in. AI drafts a response. Human approves. Response goes out. Brass policed. That's it. No LARP required.

📧

INCOMING

Customer emails sales@xyz.com (Google Workspace)

→
IMAP/webhook

🎯

TICKET CREATED

Auto-parsed: subject, body, sender, thread ID, attachments

→
immediate

🤖

SPOTTER CALL

AI reads ticket + knowledge base, drafts response

→
queued

⏳

IN THE CHAMBER

Draft queued for human review. Edit, approve, or reject.

→
approved

✅

CLEARED HOT

Response sent via SMTP through Google Workspace

→
auto

🧹

BRASS POLICED

Ticket closed or awaiting reply. Thread tracked.

Detailed Flow

📧 Email Ingestion

Poll Google Workspace via IMAP IDLE or use Google Pub/Sub push notifications
Parse: sender, subject, body (HTML → plaintext), attachments, Message-ID, In-Reply-To
Thread detection: group by In-Reply-To / References header → same ticket
Dedup: ignore auto-replies, out-of-office, own outbound
Store raw email + parsed version

🤖 AI Draft (Spotter Call)

Trigger: new ticket or new reply on existing ticket
Context window: ticket history + knowledge base articles + customer profile
System prompt: product-specific (parts catalog, lead times, compliance, specs)
AI provider: configurable — Claude API, OpenAI API, or local LLM (Ollama)
Output: draft response + confidence score + suggested category/priority
Draft is NEVER auto-sent — always queued for approval

⏳ Approval Queue (The Chamber)

Side-by-side view: original email | AI draft
Inline editing: operator can modify the draft before approving
Actions: Cleared Hot (send as-is), Edit & Send, Misfire (reject + write manual reply), Reassign
Optional: auto-approve for low-risk categories (FAQ, order status) after trust is built
Audit trail: who approved, what was edited, when

📤 Outbound & Tracking

Send via Google Workspace SMTP (appears from sales@xyz.com)
Maintain thread: set In-Reply-To and References headers correctly
Track: open status, follow-up replies auto-attach to ticket
SLA tracking: time-to-first-response, time-to-resolution
Notifications: Slack/email alerts for DANGER CLOSE tickets

Edge Cases & Safety

Misfire Handling

Rejected AI drafts are logged for model improvement
If AI confidence is below threshold → skip draft, flag for manual response
ITAR/export control keywords → auto-flag, no AI draft, human only
Profanity/threats in inbound → alert team immediately

Thread Intelligence

Customer replies to sent response → ticket reopens automatically
AI sees full conversation history for context
Auto-merge: same customer, similar subject within 24h → same ticket
Customer profile builds over time: past orders, past issues, preferences

Section 03

TECHNICAL PLAN

Self-hosted on Ragnar. Docker-composed. Node.js backend because that's what we run. PostgreSQL because tickets deserve a real database. AI provider is pluggable.

Architecture

┌─────────────────┐ ┌──────────────────┐ ┌─────────────────┐ │ Google Workspace│ │ Forward Assist │ │ AI Provider │ │ sales@xyz.com │──IMAP──│ │──API──│ │ │ │◄─SMTP──│ Node.js/Express│ │ Claude / OpenAI │ └─────────────────┘ │ + Bull queues │ │ / Ollama (local)│ │ │ └─────────────────┘ └────────┬─────────┘ │ ┌─────────────┼─────────────┐ │ │ │ ┌──────┴──────┐ ┌──────┴──────┐ ┌──────┴──────┐ │ PostgreSQL │ │ React │ │ Redis │ │ (tickets, │ │ Frontend │ │ (sessions, │ │ emails, │ │ (dashboard)│ │ queues) │ │ users) │ │ │ │ │ └─────────────┘ └─────────────┘ └─────────────┘ All containers on Ragnar (192.168.1.107) via Docker Compose Reverse proxy through existing Nginx Proxy Manager on HAOS

⚙️ Backend

Node.js + Express — REST API
Bull (Redis-backed) — job queues for email polling, AI drafting, sending
node-imap or imapflow — IMAP client for Google Workspace
nodemailer — SMTP outbound
mailparser — email parsing (HTML → text, attachments)
passport.js — local auth + optional Google OAuth
socket.io — real-time ticket updates in dashboard

🗄️ Database

PostgreSQL 16 — primary data store
Prisma ORM — schema management + migrations
Tables: users, tickets, messages, ai_drafts, approvals, knowledge_base, email_accounts
Full-text search on ticket subject + body (pg_trgm)
Redis — sessions, Bull queues, rate limiting, real-time pub/sub

🎨 Frontend

React + Vite — SPA dashboard
TailwindCSS — utility-first styling with custom "Armory" theme
React Router — client-side routing
TanStack Query — server state management
Socket.io client — live ticket updates
Dark mode only — this is the way

🤖 AI Layer

Pluggable provider interface — swap without code changes
Claude API (recommended) — best at nuanced, professional responses
OpenAI API — GPT-4o as fallback
Ollama (local) — for air-gapped / cost-conscious setups
Knowledge base: product catalog, FAQ, past responses fed as context
Confidence scoring: low confidence = skip auto-draft

🐳 Infrastructure

Docker Compose on Ragnar (Proxmox LXC or VM)
Containers: app, postgres, redis, (optional: ollama)
Reverse proxy: Nginx Proxy Manager on HAOS (192.168.1.140)
Domain: tickets.xyz.com or fa.xyz.com
SSL via Let's Encrypt through NPM
Backups: pg_dump cron → local + offsite
Git repo: git@git.jfamily.io:eric/forward-assist.git

👥 Multi-User & Roles

Admin — full access, manage users, configure AI, manage knowledge base
Operator — view tickets, approve/reject AI drafts, send replies
Viewer — read-only dashboard access (for managers/owners)
Ticket assignment — round-robin or manual
Activity log — who did what, when
Per-user notification preferences

Core Data Model

users tickets messages ├── id (uuid) ├── id (uuid) ├── id (uuid) ├── email ├── ticket_number (FA-001) ├── ticket_id → tickets ├── name ├── subject ├── direction (inbound/outbound) ├── role (admin/op/viewer) ├── status ├── from_email ├── password_hash ├── priority ├── body_text └── created_at ├── assigned_to → users ├── body_html ├── customer_email ├── message_id (email) ai_drafts ├── customer_name ├── in_reply_to ├── id (uuid) ├── created_at ├── attachments (jsonb) ├── ticket_id → tickets ├── updated_at └── created_at ├── message_id → messages ├── resolved_at ├── draft_text └── sla_due_at knowledge_base ├── confidence ├── id (uuid) ├── provider (claude/gpt) approvals ├── title ├── model ├── id (uuid) ├── content ├── status (pending/ ├── ai_draft_id → ai_drafts ├── category │ approved/rejected) ├── user_id → users ├── embedding (vector) ├── approved_by → users ├── action (approve/reject) └── updated_at ├── approved_at ├── edited_text └── created_at └── created_at email_accounts ├── id (uuid) ├── email_address ├── imap_host / smtp_host ├── credentials (encrypted) └── polling_interval

Build Phases

🔫 Phase 1 — "First Round Downrange"

Docker Compose setup (Node + Postgres + Redis)
Email ingestion: IMAP polling → ticket creation
Basic dashboard: ticket list, detail view, manual reply
User auth (local, 2-3 users)
Thread tracking (replies attach to existing tickets)
SMTP outbound through Google Workspace

Goal: Working ticket system. No AI yet. Prove the plumbing works.

🤖 Phase 2 — "Spotter Online"

AI integration: pluggable provider (Claude/OpenAI/Ollama)
Auto-draft on new tickets + new replies
Approval queue UI (side-by-side, edit, approve/reject)
Knowledge base CRUD + vector search for context
Confidence scoring + auto-skip for low confidence
Audit trail for all AI actions

Goal: AI drafts responses. Humans stay in the loop. Trust builds over time.

💅 Phase 3 — "Full Send"

Full branding pass — Armory theme, all the fun UI language
Real-time updates (Socket.io)
SLA tracking + dashboards + metrics
Notification system (Slack, email alerts)
Customer profiles + history
Auto-approve rules for trusted categories

Goal: Production-ready. Looks amazing. Your buddy actually wants to use it.

🚀 Phase 4 — "Sustained Fire" (Future)

Multi-inbox support (sales@, support@, returns@)
ITAR/compliance flagging
Canned responses library
Customer-facing portal ("check my ticket status")
Reporting: response times, AI accuracy, volume trends
Mobile-responsive / PWA

Goal: Scale it. Maybe other shops want this too. 👀

Section 04

APP MOCKUP

What the dashboard could look like. Dark mode only. Mil-spec meets modern SaaS. The kind of app that makes Zendesk look like it was designed by someone who's never touched a rifle.

Forward Assist — The Armory

FORWARD ASSIST

ticket command center

🎯 The Armory

⏳ In The Chamber 3

📋 All Tickets

📖 The Manual

👥 Operators

🔧 Field Strip

Eric J.

ADMIN

THE ARMORY

Incoming Today

In The Chamber

Cleared Hot Today

Danger Close

Ticket	Subject	Customer	Status	Priority	Age
FA-0042	Need lead time on BCG batch order (500 units)	Mike T. — Tier 1 Defense	DANGER CLOSE	Critical	4h
FA-0041	AI draft ready: Return policy for cerakote defect	Sarah K. — Sportsman's Supply	IN THE CHAMBER	High	6h
FA-0040	AI draft ready: Barrel threading specs question	Jim R. — Personal	IN THE CHAMBER	Medium	8h
FA-0039	Invoice request for PO #7734	Dave L. — Patriot Arms LLC	IN THE CHAMBER	Low	1d
FA-0038	Confirmed: Gas block alignment specs sent	Chris M. — AR Performance	CLEARED HOT	Medium	1d
FA-0037	Wholesale pricing inquiry — dealer application	Tom B. — Liberty Firearms	INCOMING	High	2h

Approval Queue — "The Chamber"

Side-by-side: what the customer said vs. what the AI wants to say. You're the final trigger pull.

Forward Assist — In The Chamber (3 pending)

FA-0041 Return policy for cerakote defect

IN THE CHAMBER

📧 Customer Email

Hi,

I received my order of 50 buffer tubes last week and about 8 of them have cerakote flaking near the threads. These are going into builds for a local PD contract so I need them right. Can I get replacements sent out ASAP? Happy to send photos.

Thanks,
Sarah K.
Sportsman's Supply

🤖 AI Draft (Spotter Call)

CLAUDE-3.5-SONNET • CONFIDENCE: 92%

Hi Sarah,

Sorry to hear about the cerakote issue on those buffer tubes — that's not the standard we hold ourselves to, especially on a LE contract build.

Go ahead and send photos to this email and we'll get 8 replacements shipped out priority. No need to return the defective ones — toss 'em in the parts bin or send them back at our expense, your call.

I'll flag this batch for QC review on our end. If you need anything else to keep that PD build on schedule, let me know.

— [Your Name]