Vantarius

Find them first. Reach them right.

Vantarius is a local LinkedIn outreach engine that reads your Excel CRM and runs a structured multi-stage prospecting sequence, powered by a local LLM (Ollama). No API keys. No cloud. Messages written by AI, sent by you (automated).

Part of the Marius Intelligence Suite.

---

What it does

Reads your CRM and, for each lead, runs the right step based on where they are in the sequence:

Stage	When	Action
J0	New lead	Connection request + personalized note
J3	3 days after J0	Follow-up DM (context/proof)
J7	7 days after J0	Value-add DM
J10	10 days after J0	Clean closing message

Each message is generated by your local LLM using the tension signal from the CRM, then validated (length, tone, CTA) before sending.

---

Requirements

• Node.js ≥ 18
• Ollama running locally with at least one model
• A LinkedIn account (login done once via node setup.js)
• An Excel CRM (use AxioMariuS to enrich it first)

---

Setup

# 1. Install dependencies
npm install

# 2. Edit config.yaml (CRM path, model, quotas, message templates)

# 3. Install a model in Ollama
ollama pull gemma3:12b

# 4. Log in to LinkedIn (one-time)
node setup.js

---

Usage

# Run full pipeline (all stages, respects quotas)
npm start

# Preview what would happen — nothing is sent
npm run dry-run

# Process only a specific stage
npm run j0    # New leads only
npm run j3    # J+3 follow-ups only
npm run j7    # J+7 follow-ups only
npm run j10   # Closing messages only

Vantarius respects daily quotas, working hours, and deduplication automatically.

---

CRM structure

Vantarius reads from (and writes back to) your Excel file. Minimum columns:

Column	Field	Description
A	Company	Company name
B	Stage	Pipeline stage (updated by Vantarius)
C	First name	Contact first name
G	LinkedIn URL	Profile URL
H	Signal	Tension signal (filled by AxioMariuS)
I	Date contact	Date J0 was sent

Exact column positions are configured in config.yaml.

---

Configuration

Key settings in config.yaml:

llm:
  model: "gemma3:12b"
  quality_threshold: 7     # Rewrite if AI scores below this

linkedin:
  quota_per_day: 15        # Max invites per day
  delay_min: 30000         # 30s minimum between leads
  delay_max: 90000         # 90s maximum

working_hours:
  start: 9
  end: 18

messages:                  # Base templates — LLM personalizes them
  j0: "I spotted friction signals at {COMPANY}..."

---

Safety features

• Daily quota — configurable cap on invites per day
• Dedup guard — never contacts the same profile twice in a day
• Working hours — no messages outside your configured window
• Dry run — preview everything before sending a single message
• Atomic CRM save — TEMP → BACKUP → RENAME, safe on OneDrive/Dropbox

---

Project structure

vantarius/
├── config.yaml              ← Edit this
├── setup.js                 ← LinkedIn login (run once)
├── src/
│   ├── main.js              ← Orchestrator
│   ├── llm_polisher.js      ← Message generation + validation
│   ├── linkedin_sender.js   ← Connection requests + DMs
│   ├── dedup_guard.js       ← Daily deduplication
│   ├── crm_writer.js        ← Atomic Excel write
│   ├── config_loader.js     ← YAML config reader
│   └── logger.js            ← Console + file logging
└── .github/workflows/       ← CI

---

Marius Intelligence Suite

Vantarius and AxioMariuS form a two-stage pipeline:

AxioMariuS   →   Vantarius
(OSINT)          (Outreach)
Enrich CRM       Read signal → generate message → send

They share the same CRM format and LinkedIn session directory. They can be used independently or together.

---

License

MIT