DEV Community

cooper
cooper

Posted on

Stop Switching Between ChatGPT, Claude, and Gemini — I Automated It

#ai #opensource #productivity #python

I spent last week tabbing between four AI chats to answer one question. ChatGPT for the summary, Claude for the code, Gemini because it has a bigger context window, and a local model because the data was sensitive.

I thought: what if I just typed the question once and the right model answered automatically?

So I built that.

pip install nvhive
nvh "What is machine learning?"
Enter fullscreen mode Exit fullscreen mode

No API keys. No config files. Works in 10 seconds.

What It Actually Does

You type a question. nvHive figures out the rest:

  1. Classifies your task — is this code? writing? math? research?
  2. Scores every available model — on quality, speed, cost, and privacy
  3. Picks the best one — and tells you which it chose 
$ nvh "Explain Python decorators"
╭─ nemotron-small (local, free) ──────────────────────────╮
│ A decorator is a function that takes another function... │
╰──────────────────────── 0.4s · 52 tokens · $0.00 ───────╯
Enter fullscreen mode Exit fullscreen mode

Coding question? Routes to a code-optimized model. Creative writing? Different model. Sensitive data? Stays on your local machine. You never pick.

The Fun Part: Council Mode

When a decision matters, one model isn't enough:

nvh convene "Should we use Redis or Postgres for sessions?"
Enter fullscreen mode Exit fullscreen mode

This sends your question to 3-5 AI-generated expert personas. A database architect, a backend engineer, a performance specialist. Each answers independently. Then it synthesizes their responses into one answer.

Like having a board of advisors that costs $0.00.

What It Costs: Nothing

25 of the 63 models are completely free. No credit card. No trial period.

  • Have a GPU? Runs AI locally on your machine. Zero cost, fully private.
  • No GPU? Free cloud providers (Groq, GitHub Models, LLM7) work instantly.
  • Want premium models? Add an OpenAI or Anthropic key whenever you're ready. 
nvh "explain quantum computing"      # free, auto-routed
nvh safe "analyze my salary data"     # forces local only
Enter fullscreen mode Exit fullscreen mode

Unix Pipes — Fits Your Workflow 

git diff --staged | nvh pipe "Write a commit message"
cat error.log | nvh pipe "What caused this crash?"
curl api.example.com/data | nvh pipe "Summarize this JSON"
Enter fullscreen mode Exit fullscreen mode

Reads from stdin. Clean text output. Composable with everything.

Web Dashboard 

nvh webui
Enter fullscreen mode Exit fullscreen mode

Dark-themed dashboard with real-time streaming, provider health monitoring, and usage analytics. Optional — the CLI does everything.

Works With Tools You Already Use 

# Claude Code
claude mcp add nvhive nvh mcp

# Any OpenAI SDK app — one line change
client = OpenAI(base_url="http://localhost:8000/v1/proxy")
Enter fullscreen mode Exit fullscreen mode

Also integrates with Cursor, NVIDIA NemoClaw, and OpenClaw.

Try It 

pip install nvhive
nvh "hello"                                # your first query
nvh tour                                   # 90-second interactive demo
nvh convene "best database for my startup" # council mode
nvh webui                                  # web dashboard
Enter fullscreen mode Exit fullscreen mode

22 providers. 63 models. 25 free. 217 tests. MIT licensed. Solo built.

GitHub: github.com/thatcooperguy/nvHive
PyPI: pypi.org/project/nvhive

Happy to answer any questions in the comments.

Top comments (0)