One command to install a friendly network-agent gateway. Chat across 23 providers. Optionally join the hivv bee swarm and earn compute credits.
curl -sSL https://hivv.org/install-hivv.sh | bash
curl -sSL https://hivv.org/install-hivv.sh | bash -s -- --with-llm --yes
Detects your hardware (CPU cores / RAM / GPU), installs llama.cpp (handles many concurrent requests, low memory), and downloads the right model:
qwen3:8b-q4 for NVIDIA 4â12 GB VRAM, qwen3:14b for âĨ12 GB,
gemma4:e2b for 8-core / no-GPU, qwen3:1.7b for 4â7 cores, qwen3:0.6b for tiny boxes.
Re-run with --no-llm to skip.
A stdlib-only Python CLI (no pip install) that lives at ~/.local/lib/hivv/hivv-cli.py with a hivv launcher in your PATH. It auto-updates on every hivv start if a newer version ships.
Anthropic, OpenAI, Google, Mistral, Groq, DeepSeek, OpenRouter, Qwen, Together, plus 14 self-hosted backends (Ollama, llama.cpp, vLLM, LM Studio, KoboldCPP, LocalAI, âĻ).
hivv start registers your machine as a bee. Heartbeats every 60s with CPU/MEM/disk/GPU metrics. You choose to volunteer or not.
Your bee id is mixed with your machine id (Linux/macOS/Windows). A copied install won't impersonate â the hub sees a new HW id.
Just Python 3.7+. No pip packages, no Docker, no Node, no build tools. The whole CLI is ~27 KB.
hivv health â ping the hubhivv chat hello there â stream a replyhivv completions MSG... â OpenAI-style JSONhivv metrics --local â your local metricshivv nodes â list all bees onlinehivv whoami â your session + hw idhivv start ¡ stop ¡ restarthivv status ¡ logs [-f]hivv info ¡ doctor ¡ speedhivv free ¡ pct --watch=2hivv update ¡ upgradehivv uninstall ¡ testhivv hi â first-run greeterhivv up â quick 3-line status checkhivv thanks â one-line warm replyhivv panic â 4-step emergency helphivv support "question" â AI diagnosishivv volunteer on â opt in to swarm jobshivv llm status â local-model helper# confirm the install hivv --help hivv up # quick 3-line status check # stream a chat reply (any model) hivv chat "explain transformers in 3 sentences" # local LLM is auto-offered during install â or force it: hivv chat "hello" --local # use local model (needs --with-llm during install) hivv llm suggest # hardware-tiered model recommendation hivv llm suggest --json # machine-readable, for scripts hivv llm status # detect + bench + suggest in one shot # join the bee swarm â heartbeats every 60s hivv start hivv status # pid, uptime, hub health, last-heartbeat hivv logs -f # tail the agent log hivv stop # one-shot health check hivv doctor # pasteable support bundle hivv speed # latency + throughput benchmark hivv test # doctor + speed + a remote chat
When you run hivv start, the agent:
POSTs /api/nodes/register with a stable id derived from your hostname + machine-id. Server returns a node_id.
Real CPU, memory, disk, and (if NVIDIA) GPU metrics. Hub tracks your bee in the swarm.
Hub can ask you to update, gracefully shut down, or (Phase 2) run a job. Works through closed firewalls â you initiate, hub responds.
On update_available message, the agent atomically replaces hivv-cli.py, logs the version, and re-execs. Zero downtime.
You're in control of every byte that leaves your machine:
hivv volunteer on to run other people's jobs.api_key=, password=, Authorization: Bearer, JWTs, and AWS/GitHub/Slack tokens are stripped before logs ship to disk or chat replies.~/.config/hivv/auth.json with mode 0600. HMAC-SHA256 signed requests with 5-min replay window.GET /api/health ¡ GET /api/public/status ¡ POST /api/chatPOST /api/auth/register ¡ POST /api/auth/login ¡ session cookiePOST /api/nodes/register ¡ POST /api/nodes/<id>/heartbeat ¡ GET /api/agents/<id>/inbox# remove hivv CLI and service (keeps config + logs) curl -sSL https://hivv.org/install-hivv.sh | bash -s -- --uninstall # also remove config and state rm -rf ~/.config/hivv ~/.local/share/hivv