Skip to content

originaonxi/aros-agent

Repository files navigation

AROS — Autonomous Revenue Operating System

One of two permanent intelligence cores at Aonxi. AROS finds new revenue. ARIA finds capital. Both learn from the same brain. Both compound forever.

Status PKM Brain Cost

Read the full Aonxi vision


Real numbers — October 2025 to March 2026

$199,000    gross volume collected
$8,000      peak single day
$2.9M       ARR velocity (that day annualized)
$0.50       cost per day to run
0           sales reps
40          customers — home care, cleaning, franchise

What AROS does

AROS finds businesses that need new customers, detects real-time signals about their situation, profiles their psychological defense mode, writes a message calibrated to bypass that defense, and sends it.

Autonomously. Every day. At $0.50/day.

Prospect → Signal Detection → ICP Score → PKM Profile → Claude Write → Send
    ↓                                                                    ↓
 SerpAPI                                                            Gmail SMTP
    ↓                                                                    ↓
 hiring?                                                          reply tracked
 expanding?                                                       outcome stored
 struggling?                                                      pattern learned
    ↓                                                                    ↓
    └────────────────── Airtable (the brain) ────────────────────────────┘

No sales team. No CRM workflow. No human in the loop.


The PKM layer — why we profile before we write

Most outreach fails not because the product is wrong. It fails because the message triggers a defense before the person reads the second sentence.

Every prospect is running one of ten defense modes. PKM Analyzertry it live (free, no key needed).

Mode Who Bypass strategy
MOTIVE_INFERENCE VCs, ex-GTM execs PURE_DATA — open with a number, never "I"
TACTIC_RECOGNITION Repeat founders, serial operators SIGNAL_HOOK — reference what they did last week
OVERLOAD_AVOIDANCE SMB CEOs, home care owners ULTRA_SHORT — under 60 words, one ask, one time
SOCIAL_PROOF_SKEPTICISM CTOs, engineers, technical leads CREDIBILITY_FIRST — exact numbers, public proof only
AUTHORITY_DEFERENCE Directors, mid-level managers PEER_PROOF — name their peers who already use it
LOSS_AVERSION Risk-aware buyers, regulated industries COST_OF_INACTION — frame what they lose, not what they gain
IDENTITY_THREAT Self-made founders, proud operators AMPLIFICATION — amplify their judgment, never replace it
TIMING_SKEPTICISM Perpetually "not ready" buyers TRIGGER_EVENT — name why now is different
COMPLEXITY_FEAR Non-technical owners burned by software SIMPLICITY_PROOF — show the one action, hide the rest
PRICE_ANCHORING Budget-constrained buyers, procurement ROI_INVERSION — cost per outcome, not cost per month

Before Claude writes a single word, AROS detects the defense mode, bans the phrases that trigger it, and writes using the specific bypass strategy for that specific person.

Before PKM:

"Hi Brian — hope this finds you well. Would love to set up a quick call."

After PKM (OVERLOAD_AVOIDANCE detected — 60-word cap, no fluff):

"Brian — 3 home care agencies in New York added 11 clients last month using Aonxi. You are hiring caregivers right now. 15 minutes Tuesday 10am?"

Same product. Different message. Every time.

Based on Friestad & Wright (1994) — 30 years of persuasion psychology.


The brain — Airtable

Every action AROS takes is recorded in Airtable.

Not just what happened. What worked. What did not. Why.

  • What defense mode a home care owner in Arizona was running on a Tuesday morning
  • What bypass strategy converted a franchise operator who had been pitched 60 times that week
  • What signal — a job posting, a new location, a review spike — preceded a close

This pattern library compounds every day. The agent running today is smarter than last month. The one running next month will be smarter than today.


Training data

AROS scoring is trained on 4 months of real Aonxi customer outcomes (October 2025 – January 2026). No synthetic data. No guesses.

Segment Customers Avg LTV Avg Duration
Retained (4+ months) 20% $5,769 4.0 months
Medium (2-3 months) 42.5% $4,910 2.5 months
Churned (0-1 months) 37.5% $1,200 0.7 months

Key finding: Industry fit predicts LTV better than any other variable. Home care and cleaning services stay 4x longer and generate 7x more revenue than off-niche prospects. This finding drives every scoring weight.


Architecture

aros.py           CLI entry point — one command runs the pipeline
scorer.py         ICP scoring (trained on real churn data)
harvester.py      SerpAPI signal detection (hiring/expanding/struggling)
writer.py         Claude Haiku email generation (PKM layer applied)
runner.py         Pipeline execution + Gmail SMTP
briefing.py       Daily owner email briefing
config.py         Environment variable loader

Stack: Claude Haiku / SerpAPI / Gmail SMTP / SQLite / Airtable


Run locally

git clone https://github.com/originaonxi/aros-agent
cd aros-agent
pip install -r requirements.txt
cp .env.example .env
# Add your keys to .env
python aros.py

Related

Repo What it does
ARIA Finds capital — the other intelligence core
PKM Analyzer Defense profiling — try it live
VISION.md The full Aonxi vision — AGI revenue layer for 400M businesses

Built by Anmol Samorigin@aonxi.com / originaonxi.github.io / aonxi.app

About

Autonomous revenue agent — scores prospects, researches via SerpAPI, writes cold emails via Claude, sends via SMTP. $0.50/day vs Gong $100K/year.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors