msanchezworld/compounder — explained in plain English
Analysis updated 2026-05-18
Diagnose which step of a sales funnel is losing the most potential customers.
Test a new offer or call to action safely using Stripe test mode before any real charges happen.
Keep a permanent log of what each growth experiment tried and what it learned for next time.
| msanchezworld/compounder | abhay-pratapsingh-ctrl/chaptr | abhishek-akkal/finova | |
|---|---|---|---|
| Stars | 0 | 0 | 0 |
| Language | JavaScript | JavaScript | JavaScript |
| Setup difficulty | hard | hard | easy |
| Complexity | 3/5 | 5/5 | 1/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Full live demo needs Brev sandbox access plus Hermes, Nemotron, and Stripe API keys, the local Python prototype only needs python3.
Compounder is an AI growth agent built to find the biggest leak in a sales funnel, pick the next test to try, run that test with safety limits in place, and write down what it learned so the next experiment can build on the last one. Instead of giving one off suggestions and moving on, it keeps a permanent record called a Learning Ledger, and every run produces a receipt showing what it observed, what it changed, which outside tools it called, what it learned, and what to try next. The demo follows a simple loop. First it looks at funnel numbers like visitors, leads, buyers, and revenue. Then it figures out where the biggest drop off is happening, chooses one test to run next, and carries out that test under guardrails that prevent unsafe actions. It proves any money related step using Stripe's test mode rather than real charges, then saves what it learned so that lesson feeds into the following experiment. In the sample case shown in the demo, 1,000 page views produced 82 leads and 8 buyers before the agent made a change, and 158 leads and 30 buyers after it tested a more specific offer. Behind the demo, an admin script starts a cloud sandbox environment, calls two outside language models, Hermes and NVIDIA Nemotron, runs Hermes inside a sandboxed tool that logs which actions were allowed or blocked, creates a test Stripe payment, and then shuts the sandbox down again. There is also a separate, simpler local prototype written in Python that simulates the same growth loop on your own machine, and it will not spend any simulated compute unless you explicitly approve it with a command line flag. The project is organized into folders for the static demo page, the recorded proof of a live run, a proxy service that connects the demo to the outside AI models and Stripe, the admin test script, the local prototype, and draft submission materials like a tweet and video script. The README is clear that everything shown publicly uses test data and sample scenarios only. It does not touch real customer records, send real charges, contact real prospects, or take any production action, and anything beyond that test scope needs separate approval first.
An AI agent that finds the weakest point in a sales funnel, tests one fix at a time under safety guardrails, and keeps a permanent record of what it learned from each test.
Mainly JavaScript. The stack also includes JavaScript, Python, Node.js.
No license information is stated in the README.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.