mahout — the ElephantPool host agent

The small open-source agent that serves AI models from your machine — and gets you paid for every token it serves. Named after the person who rides the elephant.

Linux x86_64NVIDIA & AMDRust, single binary ≈5 MBOpen source

Install

curl -fsSL https://elephantpool.ai/install-mahout.sh | sh

The installer downloads the binary, verifies its SHA-256 checksum, and puts mahout on your PATH. Nothing runs until you start it.

Platforms: Linux x86_64 today. Windows and macOS are on the roadmap — the code is portable and the engine adapters are being qualified per platform; this page will carry their one-liners the day the binaries ship.

Connect your machine

1. Sign in at app.elephantpool.ai → Machines → Generate link code.

2. On the machine:

mahout up --code <YOUR-CODE> --endpoint https://app.elephantpool.ai/api/hosts

mahout probes your hardware, runs an honest benchmark, and starts serving work your machine can genuinely handle. It auto-pauses while you're using your PC.

Verify the download yourself

curl -fsSLO https://elephantpool.ai/dl/mahout-linux-x86_64
curl -fsSL https://elephantpool.ai/dl/mahout-linux-x86_64.sha256 | sha256sum -c

Prefer reading before running? The installer script is ~40 lines of plain shell, versioned in the mahout repo and published by the same CI job that builds the binary it verifies.

What mahout does — and doesn't do

  • Serves open-weight models through a local inference engine; all downloads (weights, engine) are pinned and checksum-verified.
  • Signs a receipt for every job with a key generated on your machine — your earnings are verifiable, not declared.
  • Prompts stay in RAM, are never written to disk, never logged, never used for training.
  • Auto-pauses when your session is active; mahout pause and mahout uninstall do exactly what they say.
  • No inbound ports needed — mahout polls outward, so it works behind any home NAT.

Source code & deep dive

mahout is open source (Apache-2.0): gitlab.debuck.info/elephantpool-oss/mahout — issues and merge requests welcome. Read the full tour on the blog, or go deeper with the technical spec.

Model catalogue · Knowledge base · Blog · Sign in