Download Ollama Herd

Install

Pick one method and run it on every machine you want in the herd. All three install the same two commands: herd (the router) and herd-node (the agent on each machine).

pip install ollama-herd

pipx install ollama-herd

uv tool install ollama-herd

curl -LsSf https://astral.sh/uv/install.sh | sh

irm https://astral.sh/uv/install.ps1 | iex

brew tap geeks-accelerator/ollama-herd
brew trust geeks-accelerator/ollama-herd
brew install ollama-herd

Optional: embeddings

The embedding extra adds vision embeddings (DINOv2, SigLIP, CLIP) and a built-in text-embedding server for nomic-embed-text, which keeps embedding requests out of Ollama’s queue. It adds about 300 MB.

pip install "ollama-herd[embedding]"

uv tool install "ollama-herd[embedding]"

On Apple Silicon Macs, Ollama Herd can also route image generation (mflux, DiffusionKit), speech-to-text, and an MLX backend. Those are macOS-only and switch off cleanly everywhere else. Core routing is identical on every platform.

Next steps

1. Check the install. herd --help should print the router’s options. To see the version: pip show ollama-herd, uv tool list, or brew list --versions ollama-herd. (herd --version arrives in the next release.)
2. Start the router on your most powerful machine:
   Copy

   herd

3. Start a node on every machine, including the router’s. It finds the router on your network automatically (mDNS):
   Copy

   herd-node

   If your network blocks mDNS (common on corporate Wi-Fi), point it at the router directly: herd-node --router-url http://<router-ip>:11435.
4. Open the dashboard at http://<router-ip>:11435/dashboard and watch your machines join.
5. Point your tools at it. Any client that speaks the OpenAI, Ollama, or Anthropic API can use http://<router-ip>:11435 as its base URL.

The 60-second quickstart walks through all of this with a first request.

Troubleshooting

“herd: command not found”

The install worked, but its folder isn’t on your PATH yet. Fix it for your install method, then open a new terminal:

• uv: uv tool update-shell
• pipx: pipx ensurepath
• pip on Windows: add Python’s Scripts folder to PATH (the installer has an “Add Python to PATH” option), or reinstall with uv.
• pip on macOS or Linux: pip install --user puts commands in ~/.local/bin; add it to PATH, or use uv or pipx.

For problems after it starts, see troubleshooting.

Upgrade

Then restart herd and herd-node. Upgrade every machine: the dashboard flags nodes running a different version from the router.

Uninstall

Uninstalling leaves your data in place: logs, request traces (latency.db), and settings (env) in ~/.fleet-manager/. Delete that folder for a completely clean removal. Ollama and your models aren’t touched.

Where it’s published

Open source and private

Ollama Herd is MIT-licensed, with no account and no cloud service. Your prompts and responses stay on your network. The router sends one anonymous usage summary a day (model names, request counts, latency), on by default; it never includes prompts, completions, or hostnames. Turn it off with FLEET_NODE_TELEMETRY=false. Every field it can send is listed on the telemetry page.

Installing with an AI agent

If you use Claude Code, Codex, or OpenClaw, you can hand it this:

Install Ollama Herd on this machine with `uv tool install ollama-herd`
(or `pip install ollama-herd` if uv isn't available). Then start the
router with `herd`, start the node agent with `herd-node` in a second
terminal, and confirm http://localhost:11435/fleet/status lists this
machine. Instructions: https://ollamaherd.com/download