n3rv 0.1.0

Invisible engineering infrastructure for opencode agents.

█████████████████████████████████████████████████████████

██    ██  ███████  ████████  ██     ██
███   ██ ██     ██ ██     ██ ██     ██
████  ██        ██ ██     ██ ██     ██
██ ██ ██  ███████  ████████  ██     ██
██  ████        ██ ██   ██    ██   ██
██   ███ ██     ██ ██    ██    ██ ██
██    ██  ███████  ██     ██    ███

  Because an angel without a harness is unmanageable.
  And a harness without an angel is just expensive metal.

█████████████████████████████████████████████████████████

n3rv is an open-source engineering harness designed to contain, restrain, and orchestrate Large Language Models (LLMs). Instead of treating agents as independent chatbots, N3RV builds a rigid operational framework — handling tools, dual-store memory, and real-time project context injection. Integrates natively with opencode via MCP servers, agent skills, slash commands, and sub-agents.

Architecture • SDD Workflow • MCP Tools • Deployment • Evangelion • Security • Contributing

---

⚡ Core Architecture

The base model is the Angel: raw, uncontained, and unpredictable. N3RV is the restraint harness that binds it to your local development environment.

• MAGI Consensus Architecture: Dual-layer memory routing utilizing ChromaDB for semantic long-term retrieval and SQLite for rigid ACID-compliant session state tracking (verdicts, relations).
• LCL Project Injection: Automatic parsing of environment manifests (pyproject.toml, package.json). The agent "breathes" your project architecture from session one.
• 14 Synchronized Agent Skills (The EVAs): Specialized skill files for distinct phases of the software development lifecycle — code, testing, commits, GitHub ops, git ops, SDD (explore → propose → spec → design → tasks → apply → verify → archive), judgment-day review.
• 10 Sub-Agents: Dedicated agent configs for each SDD phase + git-ops + github-ops, isolated by responsibility.
• MCP Protocol Integration: 5 native MCP servers — n3rv-memory (semantic + relational memory), n3rv-hub (A2A task delegation), GitHub wrapper, Context7 contextual search, sequential-thinking.

---

🚀 Synchronization Sequence

Initialize the harness inside your local repository:

uv tool install git+https://github.com/juanmanueldaza/n3rv.git
cd your-project
n3rv init

Or run without installing:

uvx --from git+https://github.com/juanmanueldaza/n3rv.git n3rv init

n3rv init provisions your workspace with:

📂 Project root
├── 📄 AGENTS.md              # Coding standards with LCL-injected project context
├── 📄 opencode.json          # opencode config: MCP servers, agents, instructions
├── 📄 CONTRIBUTING.md
├── 📄 SECURITY.md
├── 📂 .opencode/
│   ├── 📂 skills/            # 14 skills (code, testing, commits, SDD phases, etc.)
│   ├── 📂 agents/            # 10 sub-agents (SDD phases + git-ops + github-ops)
│   ├── 📂 commands/          # 4 slash commands (sdd-new, judgment-day, review, handoff)
│   ├── 📂 plugins/           # lifecycle & shell-env plugins
│   └── 📂 scripts/           # MCP wrapper scripts
├── 📂 .n3rv/
│   ├── 📄 a2a-config.yaml    # Agent-to-Agent hub configuration
│   ├── 📄 skill-registry.md  # Auto-generated skill index
│   ├── 📂 memory/            # ChromaDB + SQLite (MAGI storage)
│   └── 📂 systemd/           # Background service unit
└── 📂 .githooks/
    └── 📄 pre-push            # Git pre-push hook

---

🛠️ Usage

n3rv init [--stack python|node|go|generic] [--force]
n3rv update [--dry-run] [--force-commands]
n3rv hub start                          # foreground A2A hub (development)
n3rv daemon install|start|stop|status|enable|logs   # background service
n3rv memory list|search|prune|stats     # MAGI memory operations

MCP Servers (auto-configured in opencode.json)

Server	Purpose
n3rv-memory	ChromaDB + SQLite dual-store memory
n3rv-hub	A2A task delegation (JSON-RPC 2.0)
github	GitHub API via MCP
context7	Contextual search across codebase
sequential-thinking	Chain-of-thought reasoning

---

🧠 Memory: The MAGI

Dual-store persistent memory:

• ChromaDB — Vector embeddings for semantic long-term recall
• SQLite — Relations between memories (judgments, revisions, verdicts)

Memory types: architecture, bugfix, config, decision, discovery, learning, pattern, context, summary, note.

Scopes: project (shared), session (current), personal (agent-specific).

---

📋 SDD Workflow

8-phase Spec-Driven Development pipeline, started via /sdd-new <change>:

explore → propose → spec → design → tasks → apply → verify → archive

Each phase is a dedicated skill + sub-agent. Phases write to persistent memory with topic_key: sdd-<change_id>-<phase>.

---

🗺️ Project Status

N3RV is currently under active development, built in public. We are testing restraint stability and optimizing context-delivery pipelines.

Install from source after updates:

cd ~/Projects/n3rv && git pull && uv tool install --reinstall .

---

📄 License

Distributed under the GNU General Public License v2.0 (GPL-2.0). N3RV is copyleft software — keeping the armor open ensures the angels stay contained.

---

God's in his heaven. All's right with the world.