By the Numbers
What the platform looks like under the hood right now
137
Specialized Bots
28,000+
Training Samples
250+
Database Tables
9
Microservices 24/7
50
City Pages
40
Blog Posts
Training Pipeline
How every bot gets smarter, continuously, without cloud dependency
🧠
Atlas CEO Dispatches Every 4 Hours
Atlas, the AI CEO agent, ranks all 137 bots by priority and knowledge gaps, then dispatches targeted training cycles to the GPU. No human in the loop required.
🎮
7900 XT GPU Processes Locally
All training runs on a local AMD 7900 XT. No cloud API calls, no OpenAI dependency. QwenBridge on localhost:8181 handles inference. Your data never leaves the server.
📊
Dedicated JSONL Per Bot
Each of the 137 bots has its own training dataset file. Samples are structured instruction/response pairs covering the bot's full expertise domain.
✅
Quality Scoring Gate
Every generated training sample is debug-scored 0-10. Only samples scoring 3 or higher get approved into the dataset. Low scores are queued for improvement.
🔄
Learning Feedback Loop
Implementation insights feed back into the next dispatch cycle. Sharper prompts produce better output, which produces better insights. The system self-improves.
Infrastructure
The production stack running everything right now
Operating System
Windows Server 2022
IIS reverse proxy + PM2 process manager
Primary Database
PostgreSQL 16
3 databases: billing, ecosystem, enterprise
Cache Layer
Redis
3 database slots: billing, enterprise, ecosystem
Backend Services
9 Express.js Servers
Ports 3021-3032, each with dedicated routes
P2P Federation
PeerPower Rust Crates
10 Rust workspace crates + axum registry
AI Training
AMD 7900 XT (Local)
QwenBridge inference, zero cloud dependency
Live WebSockets
4 Real-Time Services
Podcast, Diplomat, Mountain Safety, Gaming
Payment Processing
Stripe (Live)
Multi-bot checkout, webhook verification
Cost Transparency
What it actually costs to build and run this platform
Monthly Operating Costs
VPS (Windows Server 2022)
~$80/mo
Domain + SSL certificates
~$150/yr
GPU (7900 XT, one-time)
~$900
Total training tokens invested
250,000+
Cloud API costs
$0
$1
Your purchase helps sustain and grow the ecosystem. Every dollar funds new bots, better training, and improved infrastructure for everyone.
Open Protocols
PeerPower P2P is open-source. We publish what we build.
4
Protocol Specs Documented
17
Jest Integration Tests