nvHive
One curl command turns a rented Linux GPU desktop into a working AI lab.
No root. No Docker. Survives reconnects.
curl -sSL https://raw.githubusercontent.com/thatcooperguy/nvHive/main/install.sh | bash
Works on any rented NVIDIA GPU desktop — RunPod, Lambda, Vast, … — anywhere you can open a terminal on an NVIDIA machine. A few minutes later your browser opens on a dashboard where everything already works:
- A local LLM picked for your GPU's VRAM — chat and image understanding with nothing leaving the machine
- An AI Wizard that knows your workspace — reads live GPU/service state, fixes problems, RAGs over your files, searches the web
- A router across local and cloud providers — free tiers first when you have no keys, your GPU first when you do
- Storage that survives resets — models, chats, and config live on the persistent volume
The browser only opens after every service passes a health check — including a real end-to-end test where the Wizard answers a message. You pay for GPU time, not debugging time.
What's inside
Model Manager. nvHive detected your GPU at install, so every model in the catalog shows a fits-your-GPU verdict and disk size before you download. One-click install with live progress, or nvh models pull gemma3:4b from the terminal. Guide →
AI Wizard. A streaming, tool-using assistant grounded in live workspace state. It can refresh models, run safe repairs, ingest files you drag into chat (PDFs included), and cite web sources — showing cost and latency per response. Attach external MCP tool servers and their tools join its toolset.
Agent Library and council mode. 100 agent profiles across 38 categories — coding, research, creative, GPU media, ops — each mappable to a local or cloud model. Council mode runs one question through multiple models in parallel and synthesizes the answers: nvh convene "Redis or Postgres for session storage?". Cabinets →
Chat history that survives. Conversations persist server-side, browsable and resumable from every page. Pin one and it's waiting for you after a reconnect.
Studio packs. Rootless one-command installs for ComfyUI, Blender, game-dev tooling, and music production: nvh studio --install comfy -y.
Built for machines that disappear. Everything lives under NVH_HOME on the persistent volume. Downloads run as resumable jobs. nvh snapshot save / restore moves your whole state to a brand-new VM. If your persistent mount isn't auto-detected: export NVH_HOME=/mnt/persist/nvhive before installing.
Requirements
- Linux x86_64 (primary target; Windows/macOS binaries on the Releases page)
- No root, no Docker — everything installs to user-owned paths
- Python 3.11+, or none at all (
NVH_USE_BINARY=1withstart-linux.shfetches a single-file binary) - GPU optional — CPU-only machines get a small local model plus cloud free tiers
- Disk — ~2 GB for the smallest local model; the installer shows sizes and checks free space before downloading
Already have Python? pip install nvhive (extras: [vision], [rag], [all]).
If something breaks
nvh services # per-service health table
nvh services restart # recycle the stack
nvh status --deep # full diagnostic
The dashboard's Debug Report button generates a redacted report (secrets stripped) you can paste straight into an issue. Logs live under $NVH_HOME/logs/.
Commands
| Command | What it does |
|---|---|
nvh "question" |
Route to the best available model |
nvh ask "question" --local |
Local inference only — nothing leaves the machine |
nvh convene "question" |
Multi-model council with synthesis |
nvh agent run "task" |
Agentic coding with review loop |
nvh models list --all |
Fit-ranked model catalog for your GPU |
nvh services start |
Verified bring-up (Ollama → API → WebUI → smoke test) |
nvh studio --install <pack> -y |
Rootless tool-pack install |
nvh snapshot save / restore |
Move state across ephemeral VMs |
nvh setup |
Configure providers and keys |
Full reference: docs/COMMANDS.md
Documentation
| Guide | What's inside |
|---|---|
| Getting Started | Install, first five minutes, hardware, running without root, studio packs, troubleshooting |
| Models | Model Manager, GPU detection, VRAM tiers, capability matrix, nvh bench |
| Providers | Every provider, key variable, free tier and default model |
| Commands | Generated CLI reference |
| Configuration | config.yaml, env vars, NVH_HOME layout, HIVE.md, cabinets, tools, workflows |
| Web UI | The dashboard, page by page |
| Integrations | Python SDK, REST, OpenAI/Anthropic proxies, MCP, Claude Code, NemoClaw, OpenClaw, VS Code |
| Architecture | Request flow, modules, persistence |
| Testing | Running and writing tests, CI |
| Maintainers | Releasing, service order, production readiness |
| Roadmap | Plan by release, feature table, non-goals |
Notes
- Cloud providers receive the queries you route to them, under their own privacy policies. Use
nvh ask --localto keep inference local. - AI output can be wrong. Review agent-modified files before shipping them.
License
PolyForm Noncommercial 1.0.0 — use, modify, and share nvHive freely for any noncommercial purpose. Selling this code or using it commercially is not permitted. See LICENSE and NOTICE. Versions 0.40.0 and earlier were released under MIT and remain MIT.
The license does not grant rights to the nvHive name or logos; forks should use distinct names. See TRADEMARKS.
1 maintainer
Release files for nvhive 0.42.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| nvhive-0.42.0.tar.gz | 1.0 MB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| nvhive-0.42.0-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 1.8 MB
Release files / nvhive-0.42.0.tar.gz
| Download URL | nvhive-0.42.0.tar.gz |
|---|---|
| Size | 1.0 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
592362abfc0af8376e270193a0fd3fd0293f27825857216b5c9b2ba2993f25da
|
|
BLAKE2b-256 checksum How to use checksums |
d1dc9472e96296d4e4686e81f7a19cf8c0717ec76c0bfe488248385df5a1d6bf
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Sep 2, 2026.
Transparency logRelease files / nvhive-0.42.0-py3-none-any.whl
| Download URL | nvhive-0.42.0-py3-none-any.whl |
|---|---|
| Size | 806.0 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
11ba0a5c8f157a92c08fc11ac83fd21adf52e398da1efde762980a87c267a1b8
|
|
BLAKE2b-256 checksum How to use checksums |
cc2aee3094888e479b04356ad56c323cf823a5509ed23159a1776af09fbe32d3
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Sep 2, 2026.
Transparency log
