Skip to content

Latest commit

 

History

500 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

NeXgen Engine

NeXgen Engine — AI Operating Layer

CI Latest version GitHub stars License Python 3.11+ Linux and Windows

🇮🇹 Leggi in italiano · Quick Start · Why NeXgen? · Architecture · Changelog

One Canonical Source. Any Agent. Always in Sync.

NeXgen Engine is a deterministic AI Operating Layer that unifies instructions, tool configuration, secrets, and version-controlled memory across Claude Code, Codex, OpenCode, and Antigravity.

Instead of letting individual agent CLI configurations diverge across machines, NeXgen maintains a single source of truth in Git, compiled into each assistant's native format and verified by automated diagnostics.


Demo

Visual proof beats architecture diagrams. The two commands below are the whole product: see the state, fix the drift.

nexgen info    # visual dashboard: engine version, runtimes aligned, vault hygiene, secrets
nexgen shell   # interactive REPL [1-7] — manage everything without opening an AI assistant
nexgen doctor  # 33+ fail-closed checks: git alignment, MCP reachability, link hygiene, permissions

nexgen info — AI Operating Layer dashboard on Windows (v2.0.4): Host, Vault, Planes & Runtimes, Modules, Security & Diagnostics
nexgen info on Windows — Host, Vault (344 notes), Planes & Runtimes, Modules and Security & Diagnostics at a glance. Run nexgen doctor for full diagnosis.


The Three Planes

NeXgen structures agent operations into three decoupled planes:

  1. Behavior: Universal operating policies, prompts, and invariant guardrails defined in AGENTS.md and symlinked into every runtime.
  2. Configuration: Abstract MCP manifests and skills compiled deterministically into each CLI's native configuration format via nexgen sync.
  3. Memory: Plain Markdown KnowledgeVault with compare-and-swap (CAS) concurrency locking, per-section updates (update_section), and automatic Git versioning.
                         ┌──────────────┐
                         │  AGENTS.md   │ ──► [ BEHAVIOR ]
                         └──────┬───────┘
                                │
   ┌────────────────────────────┼────────────────────────────┐
   │                            ▼                            │
   │                    ┌───────────────┐                    │
   │                    │ neXgen Engine │                    │
   │                    └───────┬───────┘                    │
   │                            │                            │
   ▼                            ▼                            ▼
[ CONFIGURATION ]         [ SECRETS ]                   [ MEMORY ]
MCP Manifests / Skills    Age Multi-Recipient Store     KnowledgeVault
Claude · Codex ·          Zero-Passphrase (0600)        CAS Locked Git Notes
OpenCode · Antigravity    Per-Host OAuth Slots          Link Hygiene Map

How it compares

Different tools solve different slices. Small syncers are great for copying one MCP server quickly. NeXgen targets the full operating layer when you run multiple CLIs on multiple machines and want instructions, MCP, skills, secrets and memory to stay consistent.

Capability NeXgen Engine AgentSync mcp-sync mcps-manager dotfiles-ai
MCP sync manifest yaml → native, 9 agents symlink auto-discover bundle
AGENTS.md / instructions canonical AGENTS.md + CAS symlink template
Skills lazy catalog + deps: yes
Memory vault (Markdown+Git) CAS + update_section + vault-map
Secrets age Zero-Passphrase multi-recipient 0600 + per-host OAuth
Doctor diagnostics 33+ fail-closed checks
Windows native verified + CI + dual launchers community Python Node community
Tests 400+ unit tests partial
License PolyForm Noncommercial 1.0.0 MIT MIT MIT MIT

Capabilities as of Aug 2026 — contributions and corrections welcome. If you only need a lightweight MCP copy between two CLIs, a small syncer is the faster path. If you want zero drift across instructions, MCP, skills, secrets and memory with a doctor that fails closed, NeXgen covers all five in one place.


Key Features in v2.0.4

  • Unified Python Core (nexgen_core): Pure Python implementation running natively across Linux and Windows with 400+ unit tests, eliminating shell script divergence.
  • Deterministic Modular Layer: 8-module catalog (memory, semantic-rag, firecrawl, ocr, n8n, browser, council, sync) managed deterministically with nexgen modules list and nexgen modules set.
  • Zero-Passphrase Secrets Store: Asymmetric age encryption (99-SECRETS/secrets.yaml.age) using machine-local hardware keys (0600), isolated per-host OAuth refresh token slots, and materialized secrets.env for shells and systemd services.
  • Visual CLI & Operator Shell: Built-in nexgen info visual dashboard and standalone nexgen shell interactive REPL with selectable menu actions ([1-7]), enabling complete human management without opening an AI assistant.
  • Comprehensive Multi-Runtime Alignment: First-class support for Claude Code, Codex, OpenCode, and Antigravity (including Council seat integration).
  • Fail-Closed Diagnostics (nexgen doctor): 33+ automated sanity checks validating git alignment, manifest reachability, link hygiene, token presence, and permission boundaries.

Quick Start

Option A — Installed (recommended for daily use)

# The engine is distributed from this repository (no PyPI package yet):
uv tool install git+https://github.com/matteopasseri407/NeXgen-Engine   # or: pipx install git+https://github.com/matteopasseri407/NeXgen-Engine
nexgen info
nexgen doctor

Once the maintainer registers a PyPI token or the Homebrew tap (see docs/release-packages.md), uv tool install nexgen-engine and brew install matteopasseri407/nexgen/nexgen become the shorter paths; every release also ships an sdist, a wheel and SHA256SUMS as release assets, so any installer can verify what it downloads.

Updates via nexgen update (with confirmation) and via the scheduled guard task that runs at login + every 30 min.

Option B — Cloned (recommended for hacking the engine)

git clone https://github.com/matteopasseri407/NeXgen-Engine.git ~/KnowledgeVault
cd ~/KnowledgeVault
bash install.sh --check          # Windows PowerShell: .\install.ps1 -Check

1. Bootstrap

Already done by the installer. Verify:

nexgen sync
nexgen doctor --verbose

2. Configure your environment

Open INIT.md and paste its contents into your preferred agent CLI (Claude Code, Codex, OpenCode, or Antigravity). The agent will guide you through profile selection and module setup.

3. Align and verify (anytime)

nexgen sync
nexgen doctor

4. Interactive operator shell

nexgen info
nexgen shell

Platform Support

System Status On what evidence
Linux released the platform this is developed and used on daily; the full cycle (install, alignment, doctor, grooming, council, update) runs here and in CI
Windows released verified on real hardware and in CI; full native Python execution, dual launchers, and complete CLI alignment
macOS untested shares the POSIX paths with Linux and should work, but nobody has run it end to end; treat a failure here as expected, and reporting it as useful
Assistant Status What is covered
Claude Code complete instructions, MCP connectors, skills, guardrails
Codex complete instructions, MCP connectors, skills
OpenCode complete instructions, MCP connectors, skills
Antigravity complete instructions, MCP connectors, skills, and a Council seat; the seat was unblocked on 2026-08-22 with a stateless invocation (agy --model ... --disable-slash-commands --new-project --sandbox -p ) verified live with a nonce prompt

Architecture Boundaries

  • No Lock-In: All memories and configuration are stored as human-readable Markdown and YAML in Git.
  • Deterministic Write Paths: Knowledge notes are modified exclusively via CAS hash verification to prevent race conditions.
  • Non-Invasive Execution: The engine manages configuration as code above runtime execution; it does not intercept real-time model token streams.

See docs/architecture-contract.md and docs/sync-contract.md for the full contracts.


FAQ

Can I use this commercially? PolyForm Noncommercial 1.0.0 allows free noncommercial use, modification, and self-hosted deployments. Commercial use requires a separate agreement — see COMMERCIAL.md and LICENSE. The Python packaging and CLI tooling are intended to stay MIT-compatible; the engine's orchestration layer is noncommercial by design.

How is this different from dotfiles? Dotfiles sync files. NeXgen syncs semantics: one AGENTS.md, one MCP manifest, one skills manifest — compiled to each CLI's native dialect (JSON/TOML/YAML, different paths on Linux vs Windows), with drift detection and fail-closed guardrails. A symlink farm cannot do that.

Do I need all four CLIs? No. Install only what you use — nexgen doctor warns (not fails) for absent CLIs. Adding a runtime later is one nexgen sync.


License

PolyForm Noncommercial License 1.0.0. Free for noncommercial use, modification, and self-hosted deployments. See LICENSE for details. For commercial inquiries see COMMERCIAL.md.

About

One Git repo to configure every AI CLI — AGENTS.md, MCP, skills, memory rendered to each tool's native format, checked for drift and synced fail-closed. Linux & Windows released.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages