Thumper-Run Documentation
Local-first platform for running AI apps on your own hardware. Install apps from the catalog, manage AI models, and use opt-in network features over encrypted transport. Private end-to-end encryption remains planned.
Get Started
What is Thumper-Run?
Understand the platform, see what it automates, check your GPU compatibility.
Getting Started
Installation, quickstart guide, and system requirements.
Tutorials — Learn by Building
All Tutorials →First Image
Generate a 1024×1024 AI image with ComfyUI and SDXL.
First Chatbot
Chat with a local LLM using Open WebUI and Ollama.
First Music
Generate a 30-second music track with ACE-Step.
App Manifest
Write a .thumper.yaml to package an AI app.
Model Pack
Create .model.yaml + .pack.yaml to bundle models.
Train a LoRA
Fine-tune an image model on your art style.
Fine-Tune a Chatbot
Train a domain-specific LLM with QLoRA.
First Voice Clone
Clone a voice and generate speech with OpenVoice.
First Agent Task
Use the AI assistant to find, install, and launch an app.
Vibe Coding
Generate and edit code, workflows, and configs with the agent.
Guides
For Users
Install and run AI apps, manage models, and customize your setup.
For App Developers
Package AI tools as .thumper.yaml manifests and ship them to the catalog.
Repositories
How code and weights are stored: .thumper vs .git, ignore files, seals, branches, publishing and the API.
Editions and system map
How sync, repositories, the API and publishing fit; what the open-source node includes.
Coming from git or Hugging Face
Migration guides for git, GitHub, GitLab, Hugging Face, DVC, Oxen and lakeFS, plus a comparison.
For Model Developers
Create model and pack manifests with optimized configs for every GPU.
ComfyUI on Thumper
Models, workflows, custom nodes, and GPU-specific tips for ComfyUI.
Publishing
Submit your app to the catalog: checklist, testing, assets, and review process.
Ollama Integration
Auto-management, model pulling, API basics, and troubleshooting.
Solutions
Task-based navigation: pick what you want to do, get pointed to the right docs.
Model Caching
HuggingFace cache structure, disk management, and cache cleanup strategies.
Image Fine-Tuning
LoRA, DreamBooth, and LyCORIS training with Kohya SS — datasets, parameters, and evaluation.
LLM Fine-Tuning
QLoRA, LoRA, and DPO with LLaMA-Factory, Unsloth, and Axolotl.
Selling Fine-Tunes
Package, price, and sell LoRAs, LLM adapters, and voice models on the marketplace.
Training Hardware
GPU VRAM requirements, cost comparison, and AMD/Apple compatibility for fine-tuning.
Performance Tuning
VRAM tiers, AMD APU optimization, smart memory eviction, and GPU fallback strategies.
Agent Mode
All 32 tools, templates, model selection, approval workflow, and configuration.
Running Apps with Agent
Find, install, launch, debug, and manage apps through natural language.
Deployment Options
Desktop app, web shell, self-hosted server — compare deployment modes.
Reference
Platform Architecture
CRDT sync, offline mode, BeeKEM encryption, and self-hosting.
Troubleshooting
Symptom-based diagnosis, GPU issues, benchmarks, and agent tools.
Reference
Glossary (55 terms), FAQ (20 questions), and comparison.
File System Layout
Directory structure, external caches, disk space, and environment variables.
Comparison
Thumper-Run vs Pinokio, manual install, and cloud APIs.
Security Practices
Model safety, encryption, API keys, and vulnerability reporting.
Cost Analysis
Local vs cloud cost comparison for images, LLMs, and storage.
Environment Variables
All configurable env vars for GPU, billing, sync, and server settings.
IPC Command Reference
Complete list of 50+ Tauri IPC commands across launcher, models, sync, and auth.
Fine-Tuning Concepts
LoRA, QLoRA, DreamBooth explained — methods, cost, and when to fine-tune vs prompt.
Design Patterns
GPU fallback chains, manifest authoring patterns, and anti-patterns to avoid.
How the Agent Works
ReAct reasoning loop, tool dispatch, LLM backends, prompt assembly, and architecture.