Tier pricing: Worker $50, Swarm $100, Colony $250
DECISION
Final tier pricing locked: Worker $50/mo ($480/yr), Swarm $100/mo ($960/yr), Colony $250/mo. Annual = 20% off via coupon. Hive = enterprise/custom. No other tiers exist.
BeezKeyz subscription tiers
DECISION
Starter $10/4.5M tokens, Builder $25/12M, Growth $55/30M, Scale $100/60M, Unlimited $200/mo. Monthly reset, no rollover. No add-on packs — upgrade tier or pay for a new month.
9-person leadership team lineup locked
DECISION
Beez (Team Lead), Charlie (Client Relationships), Hana (People and Talent), Ivy (Brand and Content), Frankie (Finance), Zara (Sales and Revenue), Winston (Client Success), Flynn (Strategy and Roadmap), Hutch (Hive Keeper).
Candy does not exist — Charlie owns sequences interim
DECISION
Candy is not a real agent. Charlie owns sequences as interim owner. When a dedicated sequences agent is built, Charlie hands off. No Candy references anywhere in code or copy.
"Hire Beez" framing locked
DECISION
You hire ONE person (Beez). She brings a 9-person leadership team. Hana deploys AI specialists (up to 400+) on demand. Never say "hire 9 bees." One hire — she brings the hive.
Zara = CRO, owns Sales Pipeline nav item
DECISION
Zara owns the revenue layer: deals, pipeline, sequences, win/loss, forecasting. Zara has her own dedicated left nav item "Sales Pipeline" — NOT inside the CRM flyout. Charlie owns CRM mechanics.
Agent count displays round DOWN to nearest 25
DECISION
All public-facing agent counts round DOWN to nearest multiple of 25. 413 agents displays as "400+". Applies to all marketing copy, landing pages, and in-product displays.
Agent model map standardized
DECISION
Beez = Opus (onboarding) / Sonnet (chat). Flynn / Frankie / Hutch / Hana = Sonnet. Ivy / Winston / Zara = Haiku. Beez onboarding interview runs on Opus for ALL users regardless of tier, drops to tier routing after onboarding_completed_at is set.
Post-onboarding setup: sequential DO queue, not parallel
DECISION
Post-onboarding setup runs via sequential Durable Object queue worker — not 7 simultaneous CF Worker calls. CF Workers has a 30-second wall clock hard limit; 7 DOs with AI calls cannot complete in 30s.
Customer intelligence daily cron uses CF Workers AI free tier
DECISION
Daily customer intelligence rebuild cron uses CF Workers AI free tier model (@cf/meta/llama-4-scout-17b-16e-instruct), not paid Haiku. Rebuild triggers: logins today > 1, OR dirty flag set.
TTS = Deepgram aura-2 — permanent voice stack
DECISION
XTTS container fully removed. Deepgram aura-2 via CF Workers AI: ~200-400ms, sub-cent per call. Voice mapping: beez=athena, charlie=orion, hana=luna, frankie=hera, ivy=stella. DO NOT revisit XTTS (1-3s latency is a dealbreaker).
3 CF AI Gateways architecture
DECISION
beezus-ai-gateway (customer BKZ calls), beezus-byok-gateway (customer BYOK keys, observability only), beezus-admin-gateway (Beezus-funded calls). CF Workers AI binding calls go direct — no gateway needed.
Beezus = system of RECORD, not system of transaction
DECISION
Beezus does NOT calculate tax, process payments, move money, file returns, or take on any regulated financial obligations. Orchestrate, record, route — never become the regulated party.
KB injection at beezus-shield, not beezus-worker
DECISION
KB context injection happens at beezus-shield level (after pass/rewrite decision, before forwarding). Agents never call the KB directly — context arrives pre-injected in the request. Two-worker architecture: KB context (sync, per-request) and aggregation (isolated, cron-only).
Agent tier limits and pack system
DECISION
Worker: 50 base (+25 pack, max 75). Swarm: 125 base (+75 across 3 packs, max 200). Colony: 250 base (4th pack triggers unlimited). Pack = 25 slots, $10/mo. Gate fires on TASK ACTIVATION, not creation.
Agent MAX_ITERATIONS locked
DECISION
All agents MAX_ITERATIONS = 15, except Beez = 40. Frontend timeouts: all agents 90s, Beez 120s. Hana was previously dynamic (2/1) and could chain 200+ agent lookups — this was critical to fix.
NOTHING TEMPORARY — all customer data to D1/R2 immediately
DECISION
Every piece of customer data hits D1 or R2 immediately. No localStorage as source of truth. No session state that dies with tab close. Every turn, every decision, every personal detail — written immediately, recoverable always.
Hana modes: Active Team vs Briefing
DECISION
If agents already exist: ACTIVE TEAM MODE — present the existing team, never offer to build a new one. If zero agents: BRIEFING MODE — propose a full team. Prevents Hana from re-selling what's already been hired.
Brand tagline and hero copy locked
DECISION
Primary tagline: "YOUR PASSION. YOUR HIVE. YOUR HONEY." Hero copy: "Stop Talking to AI. Start Hiring It." Pricing CTA: "Choose your hive" (replaces "Pick your size"). Do not modify these without an explicit decision.
Onboarding interview = 16 questions, runs on Opus
DECISION
16 questions covering: preferred name, business name, description, communication style, team size, customer source, pain points, time sink, payments, website, brand personality, brand colors, ideal customer, email/texts, agent autonomy, anything else. Beezus pays ~$3-4 in Opus tokens per user.