One assistant. Every model.
Powered by everyone.
Benji answers with the best open model for each task — on your device when it can, across the Benji mesh when it helps. Private things stay private. You never pick a model.
The browser version runs a real model locally on your machine — no account, no server, nothing sent anywhere.
Just ask Benji
No model picker, no settings. Benji reads each task and chooses the right open model for it — small and instant for a rewrite, heavyweight for hard reasoning.
Runs on the best machine
Light work runs on your own device. Bigger tasks route to the nearest mesh peer that already has a stronger model loaded. The largest models run sharded across fast clusters.
Powered by everyone
Installing Benji joins the mesh. When your Mac is idle and plugged in, spare compute serves others — and earns you credits for premium capacity in return.
Private by design
Sensitive stays local
The router classifies privacy before anything else. Passwords, documents, personal details — those prompts are pinned to your device and never reach the mesh.
See where every answer ran
Every reply carries a badge: your device, a mesh peer, or a cluster. Nothing is hidden about where your words went.
You control contribution
Share compute only while plugged in, only while idle, with hard caps on GPU, storage and bandwidth. Pause any time with one click.
The mesh grows with every install
One device runs a model locally. Ten devices route tasks to whichever peer serves them best. Hundreds form clusters that run open models no single laptop could hold. Benji turns spare compute everywhere into one assistant for everyone.
Join the mesh