Vectra AI — Autonomous CRO for SMBs

Autonomous multi-agent AI sales team for SMBs that qualifies, negotiates, and closes 24/7 — audited by a Gemini Finance Guardrail.

Comment

THE PROBLEM

99% of small businesses worldwide are locked out of global markets. They lack the resources to hire 24/7 sales teams, support staff, or automated operations infrastructure. While enterprise companies deploy AI at scale, small businesses are left behind — drowning in manual work, missing deals across time zones, and unable to compete.

THE VISION

What if every small business could deploy an autonomous AI workforce that operates like a Fortune 500 sales team — at near-zero marginal cost?

INTRODUCING: VECTRA AI

Vectra is not a chatbot. It is a multi-agent autonomous system that takes over the repetitive core of a sales team: qualifying prospects, negotiating within strict compliance rules, closing deals, logging every outcome, and learning which strategies actually win — without human intervention.

Four agents work in a glass-box workflow you can watch live:

  • 🎯 Manager Agent (Orchestrator) — the central brain that routes every lead and makes escalation decisions.
  • 🤝 SDR Agent — qualifies prospects and answers questions honestly; it refuses to hallucinate features.
  • 💰 Closer Agent — negotiates under a hard 10% discount ceiling and never volunteers discounts unprompted.
  • 🛡️ Finance Guardrail (Gemini) — audits every promise in real time; violations trigger forced self-correction.

SIGNATURE CAPABILITIES

  1. Memory-aware negotiation — CRM-native long-term memory in Google Sheets; returning leads are greeted with full context ("Last time we discussed…").
  2. Autonomous outcome detection — CLOSED / LOST / NEGOTIATING is derived deterministically from the customer's own words. The AI never grades its own deals.
  3. Adaptive personality engine — DRIVER / ANALYTICAL / AMIABLE / EXPRESSIVE detection at zero LLM cost.
  4. CRO-mode strategic BI — honest executive reports from live CRM data, including candid admissions when the sample size is too small.
  5. Zero-click execution — every deal is logged into the Google Sheets CRM autonomously.

HOW WE BUILT IT

  • DeepSeek V4 Pro (Fireworks) does 95% of reasoning; Gemini 3.5 Flash-Lite does the 5% compliance audit — a deliberate 95/5 hybrid.
  • Google Sheets is the database: Lead Pipeline, CRM Log, Lead Memory, Strategy Stats. No vector DB, no paid CRM.
  • Deterministic Python for strategy, personality, outcome detection, and stats — zero LLM cost.
  • Chainlit glass-box UI, Docker, deployed on Railway.

CHALLENGES WE RAN INTO

  • Making the guardrail fail closed without killing conversion (disambiguating refusal vs. offer).
  • Model output truncation — solved with retry logic, sentence-boundary sanitizing, and punctuation instruction.
  • Keeping the entire system on free tiers while staying production-grade.

HONEST SCOPE & COMMERCIALIZATION

Vectra is an autonomous revenue-operations layer — not a lead generator or payment processor. After a month of development, we listed it on Gumroad, Whop, and Payhip three days ago at a $9 early-bird price. Zero sales so far — normal for a 72-hour-old listing with no ad spend. We would rather report an honest day-3 listing than inflate numbers we don't have: the same philosophy as the system itself, rather lose a deal than lie to win one.

WHAT'S NEXT

Payment webhooks, an email outreach channel, per-vertical knowledge bases, and a public strategy-learning leaderboard.

Built With

  • chainlit
  • deepseek
  • docker
  • fireworks-ai
  • gemini-api
  • google-sheets-api
  • json
  • multi-agent
  • python
  • railway
  • state-machine-architecture
Share this project:

Updates