Infrastructure●Mid
AlifZetta – AI Operating System That Runs LLMs Without GPUs
CPU-only LLM inference via vGPU SIMD, but prototype status and deployment clarity unclear.
Big BrainBold Bet
padamkafle
413mo ago
Back to browse
GitHub Repository
0 starsTypeScript
AluminatiAi – Per-job GPU energy cost tracking (open source)
by AluminatiAi·Mar 2, 2026·3 points·0 comments
AI Analysis
●●SolidSolve My ProblemShip It
Per-job GPU cost breakdown where cloud bills and nvidia-smi fail to deliver.
Strengths
- •Solves real pain: nvidia-smi shows instantaneous watts, billing shows monthly totals—nothing shows which job cost what. Genuine gap.
- •5-second granularity + job tagging enables precise cost attribution across training runs, not just guesswork or retrospective estimates.
- •Lightweight Python agent with zero dependencies beyond pynvml; runs anywhere NVIDIA GPUs exist including Colab, lowering friction.
Weaknesses
- •Landing page repository linked—actual agent code not visible in provided materials; can't verify implementation depth or reliability.
- •Relies on self-reported electricity rates; no auto-detection of cloud provider pricing (AWS On-Demand, GCP, Lambda rates), limiting out-of-box accuracy.
Category
Target Audience
ML engineers and researchers running GPU training jobs who need fine-grained cost attribution.
Similar To
Weights & Biases (cost tracking module) · Paperspace Gradient (cost monitoring) · Lambda Labs billing dashboards
Post Description
I built a lightweight GPU monitoring agent that streams power metrics to a dashboard and converts watts into actual dollar costs — per job, per GPU, per day.
The problem: nvidia-smi shows you watts right now. Your cloud provider shows you a monthly total. But nothing tells you which training job was the expensive one.
AluminatiAi fills that gap. A Python agent (using NVML via pynvml) runs on your GPU machine, samples power every 5 seconds, and uploads metrics to a Next.js + Supabase dashboard. It tracks energy in kWh and converts to dollars at your electricity rate. Tag your runs with job names and you get full per-job cost attribution.
Tech stack: Python + pynvml for the agent, Next.js + Supabase for the platform, Recharts for the dashboard. Open source and free to try (30-day trial, no credit card). Works on any NVIDIA GPU — tested on A100, H100, RTX 4090, and in Google Colab.
GitHub: https://github.com/AgentMulder404/aluminatai-landing Live: https://www.aluminatiai.com
Similar Projects
Productivity●●Solid
ApplyPilot – AI Agent that applies to jobs for you
6-stage job application pipeline with browser automation—solves real friction, but legally sketchy.
Solve My ProblemWizardry
pickle-pixel
313mo ago
AI/ML●Mid
Adaptive Runtime – AI agent layer, no GPU, crash recovery
Agent runtime infra, but 0 stars and crowded with LangGraph and Temporal.
Ship ItBold Bet
StateflowsLabs
3116d ago
AI/ML●●Solid
Cogveo–Schedule AI agents to run Python on your files and email results
Scheduled AI agents with per-run cost tracking beat generic chat wrappers.
SlickSolve My Problem
thrinz
301mo ago
AI/ML●●Solid
SteerPlane – Runtime guardrails for AI agents (cost limits, loops)
One decorator caps agent costs, detects loops, logs telemetry—real guardrails for runaway LLMs.
Solve My ProblemBig Brain
vijaym2k6
123mo ago
AI/ML●●Solid
LLM-use – cost-effective LLM orchestrator for agents
Smart local‑first routing that only escalates to expensive cloud planners when necessary is the standout idea — combined with per‑run cost accounting and full Ollama offline support it solves a real operational itch. The repo is a pragmatic, CLI/TUI-focused toolkit (scraping + cache, MCP server mode) that feels useful for teams wanting a no‑friction orchestrator, but it’s playing in a crowded space of agent frameworks so the novelty is incremental rather than revolutionary.
Niche GemBig Brain
justvugg
213mo ago