Free BUENATURA AI Agent Template

Free Open Source — v1.2.0

The AI agent scaffold built for sovereign operations.

Eight ready agents. A persistent knowledge wiki. Local inference at zero cost. Built on 13 principles. Works with any LLM that reads Markdown.

Use this template What is it?
8 Ready agents
13 Core principles
0 Cost per token (local)
Free Forever on GitHub

What is it

A scaffold, not a framework.

The BUENATURA AI Agent Template is a free GitHub template that gives you the exact folder structure, routing logic, agent definitions, and workflow patterns needed to run AI agents that actually work — across any LLM.

Built for one-person operations and lean teams who want serious AI capability without accumulating technical debt. Every file has a job. Nothing is there to impress.

Fork it, rename it, ship. The principles travel with you.

The 5 core principles

  • 1
    Carbon Fiber
    Maximum strength, minimum weight. Every line earns its place.
  • 2
    YAGNI
    Add complexity only when requirements demand it. Don't build for imaginary futures.
  • 3
    AI-First
    Every output comprehensible in one context window. No mystery code.
  • 4
    Uncle Bob
    One function, one job, self-documenting. Code reads like well-written prose.
  • 5
    No Em Dashes
    Banned in all output. Precision in language.

Folder structure

Everything in its place.

Clone the repo and the architecture is already solved. Routing, memory, knowledge, outputs — each separated by concern from the first commit.

CLAUDE.md
The agent router. Every task enters here. Routes by type, invokes the right skill, keeps the agent in bounds. Works out of the box.
KNOWLEDGE/
Three-layer wiki: raw sources, LLM-maintained wiki pages, content index. The agent builds and queries its own knowledge base over time.
skills/
Reusable behaviour modules — RAG search, verification, FMEA, business design, AgentOps. Load on demand. Never load all at once.
infra/
BitNet b1.58 2B4T local inference. Runs entirely on CPU. 0.4 GB RAM. Zero cost after setup. Low-complexity tasks never leave your machine.

The 8 agents

Call by command. Each does one thing.

Each agent has a defined role, scope, and skill set. Invoke by command prefix. No ambiguity about what runs or why.

@researcher
Research Agent
Gathers, synthesises, and verifies information. Uses RAG over KNOWLEDGE/ and external sources.
@comms
Comms Agent
Drafts and validates communication output via the FATE framework. Tone-tested before sending.
@decider
Decision Agent
Structures decisions using OODA. Forces explicit criteria before any recommendation.
@reviewer
Quality Reviewer
Runs guardrails and the evaluator checklist. Returns pass or fail with reasoning. No soft approvals.
@dmaic
DMAIC Agent
Executes the full DMAIC loop for recurring process defects. Define, Measure, Analyse, Improve, Control.
@project
Project Agent
Runs the 6-phase project lifecycle. Scoping through delivery with status and output tracking.
@6sigma
Six Sigma Expert
Master Black Belt depth: statistical analysis, Lean methodology, control plans, and FMEA.
@biz
Business Architect
One-person AI business design, MVP scoping, and agent stack planning. Strategy meets structure.

Knowledge Management

The LLM Wiki pattern.

Inspired by Andrej Karpathy's LLM Wiki specification. The agent doesn't just retrieve documents. It builds and maintains a persistent, interlinked knowledge base over time.

01
Ingest
Drop a source file into KNOWLEDGE/raw/. The agent reads it, writes a summary wiki page, and updates up to 15 related pages across the wiki.
Ingest KNOWLEDGE/raw/your-doc.md
02
Query
Ask a question. The agent reads the index, loads relevant wiki pages, synthesises an answer, and files valuable answers back into the wiki automatically.
Query: what do we know about X?
03
Lint
Periodically audit the wiki for contradictions, orphan pages, stale claims, and missing cross-references. The wiki improves over time rather than drifting.
Lint the wiki

Quick start

From zero to running agent in five steps.

1
Use the template on GitHub
Click "Use this template" on the repo. Name your project in kebab-case.
2
Open CLAUDE.md
Your agent router is ready. No edits required to start. Read it once.
3
Pick an agent
Open agent-scaffold/agents/README.md. Identify which of the 8 fit your use case today.
4
Add domain knowledge
Drop reference documents into KNOWLEDGE/raw/ and tell the agent to ingest them.
5
Run your first task
Route through CLAUDE.md. Confirm output lands in output/. You are operational.
Terminal — optional fast start via Codespaces
# open in Codespaces — free tier available
# click the badge in the README
 
# get a free NIM API key
# build.nvidia.com
 
# install NemoClaw (optional, alpha)
curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash
nemoclaw onboard
 
# optional: local inference, zero cost
bash infra/bootstrap.sh
# BitNet b1.58 — 0.4 GB RAM, CPU only
 
# start working
@researcher summarise our Q1 positioning

Get started

Sovereign infrastructure.
Free. Forever.

Open source and will remain free. Fork it, adapt it, or contribute back. Built by BUENATURA for anyone building lean, honest operations with AI.

View on GitHub Learn about BUENATURA

MIT License. No strings. Built by BUENATURA Holdings.