Prepare a self-hosted Thumper-Run environment for your team and keep infrastructure policy under your administration.

Why Self-Host?

Teams need shared AI infrastructure but cannot afford to send proprietary data to cloud providers. Self-hosting Thumper-Run gives your team:

Shared model library – download models once, share across the team
Centralized app catalog – IT controls which AI tools are available
Server-assisted sync – keep the configured relay under your administration
Audit logs – record the operations exposed by the configured deployment

Deployment Options

Thumper-Run supports two self-hosted deployment models:

Docker Compose (recommended for small teams, 2–20 people):

Single server with Docker
Includes the web shell, sync relay, and catalog service
Minimal configuration required

Kubernetes (for larger deployments):

Helm chart with autoscaling
Separate services for web, sync, and model storage
GPU node pools for shared inference

See the self-hosting guide for detailed setup instructions.

Quick Start with Docker

`bash

git clone https://github.com/thumper-ai/thumper-run

cd thumper-run

docker compose -f docker/docker-compose.yml up -d

This starts the web interface on port 3000, the sync relay on port 8443, and a PostgreSQL database. Team members point their desktop app at your server's URL.

Team Features

Shared catalog: Curate which apps your team can install. Add internal tools alongside public ones.
Model cache: Models downloaded by one team member are cached on the server. Others get them instantly.
Role-based access: Admin, Developer, and Viewer roles with different permissions.
Usage metrics: See aggregate usage statistics without compromising individual privacy.

Security Architecture

Self-hosting moves the service infrastructure and administrative trust boundary into your environment; it does not make the server zero-knowledge.

Current sync workflows are server-assisted and may expose content or keys needed to provide the service
Configure a supported OIDC provider and apply least-privilege roles
Use TLS for supported network hops and protect termination points
See the security docs for the current threat model and gated E2EE design

Hardware Recommendations

For a team of 10:

CPU: 8+ cores
RAM: 32 GB
Storage: 500 GB SSD (for models and app data)
GPU (optional): One GPU enables shared inference for team members without local GPUs

Ready to try it? Download Thumper-Run free →