Alpha release – under active development

AI agents need more than raw code—they need context, structure, and guidance.

Dokugent is a markdown-native, documentation-first CLI tool for building, testing, and coordinating AI agents using simple Markdown and JSON scaffolds. It prioritizes transparency, human-in-the-loop workflows, and model-agnostic compatibility.

Instead of bloated repos or scattered prompts, Dokugent gives your AI teammates a clean, token-efficient folder of structured instructions, ready to reuse across projects.

Dokugent also helps teams save significantly on development costs by letting them simulate agents locally using Ollama before making any API calls. You can test variations, debug behaviors, and explore edge cases without burning tokens — governance and traceability become natural side-effects of good development hygiene.

While originally designed for developers, Dokugent’s modular structure can also support structured thinking in content planning, education, research, and more.

• 📁 Scaffolds agent projects with init and wizard
• 🧠 Plans and compiles agent behavior from Markdown
• 🛡️ Enforces conventions and traceable criteria
• 🔐 Certifies and simulates agent reasoning flows
• 📦 Supports Claude, Codex, GPT-4, Gemini, LLaMA, Nvidia via agent --ecosystem or conventions

Test smarter. Spend less. Ship safer.

Dokugent helps developers design and test AI agents locally — using Ollama — before committing to expensive API calls. It’s a dev stack for agent builders who want to:

• 🏠 Run full agent simulations offline with no API cost
• 💸 Save 60–80% of LLM spend by testing with Ollama before calling expensive APIs
• 🔏 Sign, certify, and trace agent decisions with governance built-in
• 📜 Deploy with audit trails, signer identities, and Doku URIs
• ⚖️ Governance becomes an emergent property, not a tax on your workflow

Start local. Scale safely. Pay only when it matters.

npm install -g dokugent

📘 Full Setup Guide

This is a founder’s release (v0.1) designed for testing structure, simulation, and governance flows. It is not production-ready, and automated test coverage is not yet implemented.

🧪 That said:

• Core features do work
• We eat our own dogfood daily
• CLI feedback is welcome!

If you're exploring agent governance or traceability workflows, we’d love your input. Feel free to open issues or start a discussion.

• ✅ dokugent init — Scaffold a new project
• ✅ dokugent owner — Set or view project owner metadata
• ✅ dokugent agent — Create a new agent identity (--t for template)
• ✅ dokugent keygen — Create identity keypairs

• ✅ dokugent plan — Draft an agent plan
• ✅ dokugent criteria — Define evaluation criteria
• ✅ dokugent conventions — Select AI conventions
• ✅ dokugent byo — Import external agent JSON payload
• 🔲 dokugent compliance — Fill in GDPR & governance metadata
• 🔲 dokugent io — Fill in I/O & Rules

• ✅ dokugent preview — Generate agent spec bundle
• ✅ dokugent certify — Sign and freeze validated preview
• ✅ dokugent compile — Build deployable agent bundle
• ✅ dokugent deploy — Run full deploy (preview → certify → compile)

• ✅ dokugent dryrun — Simulate plan execution without real actions
• ✅ dokugent inspect — Inspect agent cert or plan (local or MCP)
• ✅ dokugent security — Scan for file-level threats
• ✅ dokugent simulate — Run simulated agent logic with any LLM on your Ollama
• ✅ dokugent trace — Trace agent behavior from a dokuUri

• ✅ dokugent mcp-schema — Generate MCP-compatible JSON schema for agent plans

• ✅ dokugent audit — Verify agent project structure and check for missing or malformed files
• ✅ dokugent ethica — Simulate ethical dilemmas, persona debates, and council-based reasoning flows
• ✅ dokugent security — Scan and detect risks in agent metadata and input files
• ✅ dokugent status — Classify agent readiness across lifecycle stages
• ✅ dokugent trace — Trace agent behavior for transparency and audits
• 🔲 dokugent redteam — Stress-test agent plans with adversarial vectors

• 🔲 dokugent fetch — download community-contributed agent plans and templates
• 🔲 dokugent test — run internal CLI checks and command tests (coming after v0.1 release)

Dokugent embraces a protocol-first mindset for building intelligent systems. You don’t start by coding — you start by thinking, documenting, and aligning. This structure keeps your agents safe, traceable, and easy to reconfigure.

With Dokugent, your documentation becomes a reusable memory scaffold.

How do you pronounce Dokugent? Like Goku, but with doku — which in Japanese can mean either:

• 読 (doku) — “to read”

Add agent and you get: Dokugent = a reading agent…

• 🇯🇵 書類読解エージェント (Shorui Dokkai Eejento) → Literally: “Document Comprehension Agent” (書類 = documents, 読解 = reading comprehension)

• 🇵🇭 Dokumento → Tagalog for "document" (from Spanish documento) → Used commonly as "mga dokumento" for “documents”

• 🤖 Agent = from English, written in Japanese as エージェント (eejento)

Dokugent is a portmanteau of all these — a cross-cultural nod to literacy, power, and intelligent agents.

This project is built by a small but mighty squad:

• Carmelyne — Human brain behind it all; UX tactician, debugger, and design conscience.
• BeshLLM (ChatGPT4o) — Dev bestie sidekick, logic prompter, code stylist, & pun enabler.
• Oboe — File patcher and terminal ghost; never seen, always reliable.
• ChatGPT-4 — Occasional contributor. Brilliant. Unpredictable. May be tipsy.

Dokugent is licensed under the PolyForm Shield License 1.0.0.

This is a protective open-source license designed to encourage collaboration while preventing direct competition.

✅ You can:

• Use Dokugent for personal projects
• Use it in client work
• Integrate it into internal tooling

❌ You can’t:

• Use it to build a directly competing product or service

⚠️ TL;DR: It’s open to build with, not open to clone against.

📬 Contact Me

Have questions, feedback, or want to collaborate?

Carmelyne Thompson

• 💻 carmelyne.com
• 🧵 @carmelyne
• ✖️ @carmelyne
• 💼 linkedin.com/in/carmelyne
• 📧 Email: hello@carmelyne.com

🧠 Structured with AI Agents in mind.