OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Infrastructure & Serving

Berth

Berth is a self-hosted GPU resource broker with a Go command-line client and daemon. It queues applications that request GPU memory, manages handoffs after workloads finish or unload, and reports which processes hold or await a lease.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Berth — OpenViST project profile in AI Infrastructure & Serving.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Berth when several local applications share a GPU and need coordinated access, particularly when one workload should wait for another to release memory before starting.

Who it is for

  • People running multiple local AI inference or media applications that compete for GPU memory.
  • Operators who want a command-line tool to launch GPU programs through a broker and inspect lease status.
  • Users coordinating workloads such as chat-model servers, ComfyUI pipelines, voice engines and games.

Documented uses

  • Queue competing applications for GPU access instead of allowing simultaneous memory use.
  • Hand a GPU lease to another workload after a current reply or image job finishes.
  • Inspect which engines hold or await the GPU and identify unleased processes using VRAM.
  • Launch an inference server through Berth with a selected request priority.
Before you choose

Requirements & limits

Berth is self-hosted and distributed under the Apache-2.0 license. The README describes Linux service installation and marks Windows support experimental.

  • Berth coordinates GPU resources; it does not provide inference models or perform inference itself.
  • Clients can request access through berth run, the gateway, ComfyUI's API or a game claim.
  • The handoff process waits for the current job boundary, unloads the holder and checks that the driver reports memory free.
  • The README describes support for engines in Docker or Podman and RAM parking for engines that support the relevant offload behavior.
  • The project labels itself alpha, and Windows support is experimental.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install Berth services and check GPU leases

Install the Berth daemon and harbor as user services, then inspect the GPU and lease status.

Before you start

  • A Linux or Windows system; Windows support is marked experimental.
  • Go installation or a Berth release binary.

Steps

  1. Install the CLI with the command below, or use a release binary.

    Documented command
    go install github.com/T-Crypt/berth/cmd/berth@latest

  2. Run berth service install --harbor to install the daemon and harbor as user services.

  3. Run berth status to view the card, leases, engines and estimated VRAM.

Expected outcome

Berth services are installed for the user session and the CLI displays GPU and lease status.

Guide limitations

  • Windows support is experimental.
  • The documented status output reports estimated VRAM.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
24 October 2026
Original website title
T-Crypt/berth · Official homepage title · retained in its source language
How we review sourcesSuggest a correction