Local-first AI orchestrator

Your own AI platform. Runs on your hardware.

Puente stands up a full local AI stack — chat, image, speech, search, notebooks — from one config file. Private by default. Yours to keep.

Pay a time tax, not a token tax. Modest hardware, honest bills.

Get started → Self-host guide

$ pip install locopuente  ·  runs on a single consumer GPU

puente — bash
The cloud way

A token tax

Every prompt metered. Costs scale with use, data leaves your walls, and the bill never stops. Fast — but rented, and never yours.

The Puente way

A time tax

Slower on cheap silicon, sure. But it's a one-time cost you already paid: the machine on your desk. No metering, no egress, no surprises.

One stack, many tools

Fire up the AI services you need

Toggle services in puente.yml. Puente handles the containers, GPUs, models, and a reverse proxy — you get URLs.

Three commands

From bare machine to running platform

Puente detects your hardware, picks services that fit, and brings the stack up. No hand-wiring containers.

Detect & plan

$ puente init

Scans your CPU, RAM, and GPUs, then proposes a service set your hardware can actually run. Writes puente.yml.

Install

$ puente install

Pulls Docker images, installs native pieces (Ollama, models, ComfyUI). Pins each model to the right GPU.

Launch

$ puente up

Starts the stack and a launcher portal. Run puente status to see what's live. That's it.

Self-host

Private by architecture, not by promise

  • ✓Your data never leaves the box. Inference runs on hardware you own.
  • ✓Bring your own proxy — or use ours. Caddy ships as an optional service with automatic TLS.
  • ✓Config as code. The whole stack is one committable puente.yml.
  • ✓Runs on modest hardware. A single consumer GPU is enough to start.
# install (the command is `puente`)
$ pip install locopuente

# set it up and go
$ puente init     # detect hardware, choose services
$ puente install  # pull images + models
$ puente up       # your platform is live