NX AI Engineer

v2.0.1 · stdlib-only · zero runtime dependencies · 250 tests green · CHANGELOG · ROADMAP · RELEASE_NOTES

NX AI Engineer is a Developer Infrastructure Platform for AI-assisted software development. Its job is to organize knowledge, context and execution so that any AI model becomes dramatically more effective while building software.

NX AI Engineer does not replace Claude Code, GPT, Gemini or any other model. All intelligence belongs to the model. All organization belongs to NX.

It installs like any modern tool — a package from PyPI and a single nxai CLI — and works on any stack (Node, Python, Go, Rust, Java, Nx and other monorepos, …), which it discovers at runtime.

Guides: Installer · Upgrade · Architecture · SDK · Engine · Workflow · Plugin · Knowledge · Project Brain · Contributing

Install

pip install nx-ai-engineer
nxai version
nxai doctor

No manual file copying, ever. The platform code lives in the installed packages; your project's .ai-project-assistant/ holds data only.

Scaffold a new project (nxai new)

NX AI Engineer can also create the project — a Cloud-Agnostic production foundation in one command (the create-next-app moment for AI-native platforms):

nxai new my-platform
cd my-platform
cp .env.example .env        # Supabase URL + keys (free tier)
make up                     # boot backend + frontend via Docker Compose
nxai platform-audit         # 8-dimension production-readiness check

It generates a FastAPI backend (12-factor, structured logs, request-id, a decoupled Supabase adapter), a Next.js frontend, Docker Compose (dev/override/prod), Dockerfiles, a Makefile, multi-environment config, Supabase migrations + RLS, and docs — no business logic, just the foundation — runnable on Docker, any cloud, a VPS, Kubernetes or Swarm. See the Scaffolding guide.

Quickstart

From the root of any repository:

nxai init                      # scaffold .ai-project-assistant + audit + Brain + Knowledge + Vault
nxai plan "Add OAuth login"    # plan a goal into a task (agents, order, locks)
nxai execute "Add OAuth login" # full end-to-end flow (Dry Run -> Test -> Execute; dry-run by default)
nxai review                    # consolidated diff review
nxai knowledge status          # the three memories: Brain / Obsidian / Git
nxai docs                      # the bundled guides

Everything is safe by default: execution always runs Dry Run → Test → Execute and defaults to dry-run.

What it does

NX turns a single request into a disciplined flow — never "implement immediately":

audit → discover → impact → risk → plan → subtasks → agents → order →
implement → test → review → consolidate → document → report

It coordinates specialized agents (backend, frontend, database, ai, security, devops, qa, docs, plus read-only architect/planner/reviewer/delivery) that work safely on the same codebase via advisory locks and isolated git worktrees.

The three memories

The Knowledge Engine has exactly five responsibilities — discover, index, relate, update, deliver_context — and never reasons, learns programming, or interprets code. It only reduces the model's cognitive load: the richer the project history, the sharper the model's work, the fewer tokens spent.

The Engineering Contract

How knowledge reaches the agent. An agent doesn't get an ad-hoc prompt — it gets a contract: a declarative, predictable statement of what applies to a task.

Task → Engineering Contract → Context Builder → Model → Result → Knowledge Update

The contract assembles context (files/areas), knowledge (ADRs/patterns), engineering (the Engineering Packs that apply), constraints, requirements (mandatory tests/validations/checklists) and brain facets. Packs are contracts: they declare applies_to, so the Backend Agent automatically receives the Security, LGPD and Multi-Tenant contracts — without anyone remembering to ask. The contract is enforced at delivery (a gate blocks shipping untested code when an applicable pack mandates tests). See the Engineering Contract guide.

nxai contract --agent backend "Implement Google login" --files auth.py --areas api/auth

CLI

…plus decide, dispatch, context, run, deliver, pipeline, metrics, insights, recommend, worktree, tasks, locks, unlock, status. (The legacy nx alias is also installed.)

Engineering Packs

Domain knowledge — not code — that makes the agents apply a domain's rules correctly:

nxai pack list                 # browse the catalog
nxai pack add lgpd             # install LGPD/privacy policies + checklists + context
nxai pack add security         # OWASP/ASVS-aligned application security

Built-in packs: lgpd, security, repo-standards (stable) plus scaffolds for owasp, ai, cloud, docker, multi-tenant, observability, testing, billing, authentication. Once installed, the Pack Provider feeds the pack's policies/checklists/context to the agents working in that domain. Packs contain no code and no AI. See the Packs Guide; third parties can publish their own (the pack Marketplace).

Packs (knowledge) × Specialist Agents (execution)

The knowledge lives in the Pack; a Specialist Agent just executes using it — the agent is disposable, the knowledge is not. The Database category ships postgres and mongodb (full: rules, patterns, anti-patterns, performance, checklists, security) plus mysql/sqlserver/oracle/sqlite/redis/cassandra/elastic/ neo4j, with specialist agents database-relational, database-nosql and a read-only database-reviewer (never implements — it asks and blocks on duplicate tables, redundant indexes, anti-patterns). The Engineering Contract auto-attaches the right pack to each agent, so the same PostgreSQL knowledge serves the relational, reviewer and migration work without duplication.

nxai pack add postgres
nxai contract --agent database-relational "Add an orders table"   # postgres pack auto-attached

Standardize a repository

The repo-standards pack also ships concrete templates that nxai scaffold lays into a project's repo root — governance files, issue/PR templates and a CI workflow matched to the stack:

nxai scaffold --stack auto --dry-run   # preview
nxai scaffold --stack python           # CONTRIBUTING, CODE_OF_CONDUCT, SECURITY,
                                       # .github/ISSUE_TEMPLATE, PR template, CI, .gitignore

Idempotent — it never overwrites an existing file without --force.

Architecture

A monorepo of 9 acyclic, stdlib-only packages:

nx-core ← nx-workflow ← nx-sdk
nx-core ← nx-packs
nx-core ← nx-providers ← nx-obsidian ← nx-knowledge
{core, workflow, sdk, knowledge} ← nx-runtime ← nx-cli

See docs/ARCHITECTURE.md. The acyclic graph is enforced by scripts/verify_packages.py; the full Quality Gate is scripts/quality_gate.py.

Extending

NX is built on an Open/Closed SDK — register agents, engines, workflows, adapters, plugins and tools without touching the core. See the SDK Guide and Plugin Guide.

Principles

• All intelligence is the model's. NX organizes; it never reasons.
• Stdlib-only core — zero third-party runtime dependencies.
• Safe by default — Dry Run → Test → Execute; nothing destructive without intent.
• The Brain never stores code — only structured knowledge.
• Knowledge is leverage — more history ⇒ better context ⇒ fewer tokens.

License

MIT — see LICENSE.