-
Notifications
You must be signed in to change notification settings - Fork 0
Home
Wild_Root_Prompt turns a rough one-line idea into an expert-level prompt or a full reproducible instruction manifest, entirely on your own machine. It runs on Ollama or any OpenAI-compatible local backend, applies 173 prompt engineering techniques across 15 categories, and costs nothing — MIT licensed, no API key, no account, no subscription.
This wiki is the complete documentation. The README is the two-minute version.
See it work before you install anything: 17 words of nonsense turned into a structured technical brief →
An open-source prompt optimization tool for local large language models. You type a task in plain language; it does five things:
- Pre-processes your input — a local model restructures and completes it, turning "its slow fix it" into a real specification with an explicit deliverable.
- Injects prompt engineering techniques — 173 of them: Chain-of-Thought, Tree-of-Thought, ReAct, MECE, Constitutional AI, red teaming, pre-mortem, steelmanning and more.
-
Optionally grounds it in web context — a search that adds real-world information (toggleable, fully skippable with
--offline). - Generates with one model, or two models in parallel with split-screen live streaming.
- Synthesizes both outputs into one unified document.
You get either a Quick enhanced prompt to paste into any LLM, or a Full 12-section manifest an AI agent can execute end to end.
| I want to… | Go to |
|---|---|
| See real before/after output | Examples |
| Install it in one command | Installation |
| Get my first result in 2 minutes | Quick Start |
| Use it without a terminal | Web UI |
| See every command and flag | CLI Reference |
| Understand the 173 techniques | Prompt Engineering Techniques |
| Change tone, format and depth inline | Slash Metacommands |
| Start from a ready-made task | Templates and Bundles |
| Know what people build with it | Use Cases |
| Compare it to other approaches | Comparison |
| Tune models, temperature, backends | Configuration |
| Drive it from my own program | REST API |
| Understand the pipeline internally | Architecture |
| Look up a term | Glossary |
| Know what it doesn't do | Limitations and Roadmap |
| Fix something broken | Troubleshooting |
| Ask a quick question | FAQ |
| Contribute | Contributing |
Most prompt tools are thin wrappers: they prepend "You are an expert" and call a hosted API. Wild_Root_Prompt takes a different position on three axes.
The 173 techniques come from published prompt engineering research and practice — Chain-of-Thought, Tree-of-Thought, ReAct, MECE — categorized and selectable individually, by range, by task-type bundle, or recommended automatically from your task's own wording. A small model with 15 well-chosen techniques routinely beats the same model prompted naively, at zero extra VRAM cost.
Generation runs against a model on your own machine. Optional web enrichment is the only outbound traffic and can be turned off permanently. There are no telemetry calls, no accounts, and no API keys anywhere in the codebase — because there is nothing to authenticate to. --offline gives you a fully air-gapped run.
Parallel dual-model generation runs two different models side by side — a systematic one next to a creative one — then merges their strengths in a synthesis pass. Their disagreements are the useful part. No single-model chat interface offers this.
| Price | Free. MIT licensed, personal and commercial use |
| Account | None |
| API key | None |
| RAM | ~4 GB for a 3B model; ~8 GB for the 7–8B sweet spot |
| Platforms | Linux, macOS, Windows, Android via Termux |
| Requirements | Python 3.8+ and Ollama — both handled by the installer |
| Time to first output | About two minutes |
git clone https://github.com/TFD-42/Wild_Root_Prompt.git && cd Wild_Root_Prompt && chmod +x install.sh && ./install.shWindows, Android and double-click installs: Installation.
It improves the prompt, not the model's reasoning capacity. It structures thinking; it does not verify facts. Full manifests are slow on modest hardware, and past ~30 active techniques output quality drops. All of it, with workarounds and what's planned: Limitations and Roadmap.
- Prompt Engineering Guide — community reference for LLM prompting
- Ollama — the local model runner Wild_Root_Prompt is built on
- Chain-of-Thought Prompting — Wei et al., 2022
- Tree of Thoughts — Yao et al., 2023
- ReAct — Yao et al., 2022
- Constitutional AI — Bai et al., 2022
MIT — free for personal and commercial use.
Wild_Root_Prompt is an open-source prompt engineering tool for local LLMs: prompt optimization, prompt generation, and instruction manifest creation with Ollama, LM Studio, GPT4All and text-generation-webui. Topics: prompt engineering · local LLM · Ollama · offline AI · privacy-first AI · Chain-of-Thought · Tree-of-Thought · ReAct · MECE · AI agents · open source.
Start
Reference
Understand
Help