Skip to content

Latest commit

 

History

17 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

NEXUS

CI License: AGPL v3

Status: demo — public demo derived from G0DM0D3 (AGPL). Live · demo release

Licensed under AGPL-3.0. NEXUS is based on G0DM0D3. Local work in this repository is E2B orchestration, skills, UI, BYOK hardening, and the public demo. See ORIGINAL_WORK.md.

A multi-model AI workspace with an autonomous agent. NEXUS combines chat, file analysis, model routing, and sandboxed code execution in one interface.

Try the live demo · Read the API guide · See original work and provenance · Report a security issue

The public demo is suitable for evaluating the interface. Model and sandbox features require your own provider keys, and availability depends on those external services.

NEXUS autonomous agent analyzing data and running Python in a sandbox

NEXUS multi-model workspace welcome screen   NEXUS local-first provider settings

What it demonstrates

  • Autonomous agent workflow — plans tasks, runs Python or shell commands in an isolated E2B sandbox, and returns generated files as downloadable artifacts.
  • Skill routing — selects focused playbooks for web design, planning, data analysis, debugging, and autonomous execution.
  • Visible execution — shows a collapsible reasoning summary and a live tool timeline so users can follow progress.
  • File analysis — handles images, PDF, Word, Excel, CSV, audio/video, and common archive formats.
  • Multi-model orchestration — can race supported models or synthesize their responses into one result.
  • Bring your own provider — supports OpenRouter and NVIDIA NIM model keys.
  • Local-first settings — chat data and provider configuration are stored in the browser, with backup export/import.

Good fit for

NEXUS is useful as a reference or starting point for teams evaluating agentic workflows that must analyze files, execute code in an isolated environment, and return reviewable artifacts. It demonstrates the product flow and integration boundaries; it is not presented as a turnkey multi-tenant SaaS.

Typical client work based on this evidence includes adding an AI workflow to an existing product, connecting provider and sandbox APIs, designing visible tool execution, and hardening loading, error, cancellation, and artifact-delivery states.

Quick start

Requirements

  • Node.js 20 or newer
  • An OpenRouter or NVIDIA NIM key for model-backed features
  • An E2B API key for autonomous sandbox execution
git clone https://github.com/jc-morales-dev/NEXUS.git
cd NEXUS
npm ci
cp .env.local.example .env.local
npm run dev

Open http://localhost:3000, then add your model provider key in Settings. Add the E2B key to .env.local only when you want to exercise the agent:

E2B_API_KEY=e2b_your_key_here

Never commit .env.local or a real provider key. The file is ignored by Git; .env.local.example documents the expected variables without credentials.

Runtime modes and limitations

  • The autonomous agent needs server mode (npm run dev or npm run start) so the server-only agent route can reach the sandbox provider.
  • A static export (NEXT_STATIC_EXPORT=1 npm run build) exposes the chat UI but cannot run the agent route.
  • Browser-stored provider keys are appropriate for personal demos. A multi-user production deployment should move provider access, quotas, authentication, and audit controls behind a server.
  • Generated output and model responses must be reviewed before they are trusted or used in a production workflow.
  • CI validates lint, types, unit/security regressions, two browser journeys, and the production build. Provider calls are deterministically mocked in E2E; live third-party availability remains outside CI.

Quality checks

Every push and pull request runs all five steps (workflow):

npm run lint       # next lint
npm run typecheck  # tsc --noEmit
npm test           # stream, redaction, consent, and telemetry regressions
npm run test:e2e   # real browser journeys with deterministic provider mocks
npm run build      # Next.js production build

The regression suite pins consent, telemetry validation, credential redaction, and incomplete-stream handling. Playwright covers BYOK setup, an agent response, persistence after reload, and recovery from a failed agent request.

Architecture

src/
  app/                Next.js app and server-only /api/agent route
  components/         Chat, messages, settings, and execution UI
  lib/
    agent.ts           Agent loop, sandbox tools, and artifact capture
    skills.ts          Skill playbooks and routing
    system-prompt.ts   NEXUS agent identity and constraints
    files.ts           Client-side file extraction
    openrouter.ts      OpenRouter provider integration
    nvidia.ts          NVIDIA NIM provider integration
    tuning*.ts         Context-aware model parameters
  store/               Zustand state persisted in localStorage

E2B provides the isolated execution environment; the agent UI/loop is derived from G0DM0D3; local work centers on orchestration, skills, tool timeline, and demo hardening.

Documentation

License and attribution

NEXUS is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). See LICENSE.

The interface and several features are based on G0DM0D3 by elder-plinius, also licensed under AGPL-3.0. Because NEXUS is a derivative work, it remains under AGPL-3.0. If you distribute it or run a modified version as a network service, review the license obligations and provide the corresponding source as required. Full attribution is available in NOTICE.md.

About

Demo: multi-model AI workspace + sandboxed agent. Derived from G0DM0D3 (AGPL); E2B/skills/UI local.

Topics

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages