AI agent that transforms existing codebases — no migrations, no rewrites, directly on production code.
Cloud · Website · Issues · Discussions
Noxen understands your codebase, researches best practices, plans improvements, and implements them — with your approval at every step. You own the code. You own the API keys. Noxen provides the intelligence.
- Spider Analysis — Deep codebase scanning: languages, frameworks, services, dependency graphs, complexity metrics
- Research Agent — Autonomous discovery of best practices, patterns and solutions for your stack via GitHub, Exa, Firecrawl
- Board Deliberation — Multiple LLMs analyze your question independently, review each other's answers, then a chairman synthesizes the final response
- Execution Engine — Approved plans are executed via Claude Code in isolated sandboxes with rollback support
- Skill System — Auto-generated reusable skills from research, with confidence scoring and version control
- Knowledge Base — Qdrant-powered semantic search across all indexed codebases
- Notifications — Telegram, Slack, Google Chat, Email alerts for approvals and completions
- Web Dashboard — Real-time monitoring, chat, research, skill browser, execution viewer
┌───────────────────┐
│ Web Dashboard │
│ (localhost:8400) │
└────────┬──────────┘
│
┌──────────────┼──────────────┐
│ FastAPI Server │
│ (Orchestrator) │
└──┬────┬────┬────┬────┬──────┘
│ │ │ │ │
┌───────┘ ┌─┘ ┌─┘ ┌─┘ └───────┐
│ Board │ │Chat│ │Exec│ │Research│
│(Multi │ │SSE │ │Eng │ │ Agent │
│ LLM) │ └────┘ └────┘ └───┬────┘
└──┬────┘ │
│ ┌─────┴─────┐
┌───────┼───────┐ │ Qdrant │
GPT Claude Gemini │ (VectorDB) │
│ │ │ └───────────┘
Ollama Grok DeepSeek
git clone https://github.com/noxen-ai/noxen
cd noxen
cp .env.example .env
# Add at least one LLM API key in .env
docker compose up -d
open http://localhost:8400- Docker and Docker Compose
- Internet connection (LLM providers + license validation)
- At least one LLM API key
| Provider | Models | Type |
|---|---|---|
| Ollama | Llama 3, Mistral, CodeLlama, Qwen | Local (free) |
| Gemini 2.5 Pro, Gemini 2.5 Flash | Cloud API | |
| Anthropic | Claude Sonnet 4, Claude Opus 4 | Cloud API |
| OpenAI | GPT-4.1, o3, o4-mini | Cloud API |
| DeepSeek | DeepSeek V3, DeepSeek R1 | Cloud API |
| Grok | Grok 4 | Cloud API |
Minimum .env to get started:
NOXEN_GEMINI_API_KEY=your-key-hereFor Board mode (multi-model deliberation), configure 2+ providers:
ANTHROPIC_API_KEY=sk-ant-...
NOXEN_GEMINI_API_KEY=AI...
OPENAI_API_KEY=sk-...See .env.example for all options.
For teams and production workloads, Noxen Cloud adds:
| Feature | Description |
|---|---|
| AutoResearch | Automated research funnel — explores 100 areas, tests 20 experiments, keeps the top 5 |
| Advanced Board | Enhanced 3-stage deliberation with chairman selection, consensus scoring, history |
| Conversation History | Persistent chat with sidebar navigation and search |
| Multi-tenant | Team workspaces, per-user LLM keys, role-based access |
| OAuth + 2FA | Google/GitHub SSO, two-factor authentication |
| i18n | Full internationalization (EN/IT + more coming) |
| Usage Analytics | Token spending, model routing telemetry, cost tracking |
| API Tokens | Programmatic access for CI/CD integration |
- Free: personal use, evaluation, non-commercial projects, single-org self-hosted
- Requires license: commercial SaaS, multi-tenant deployments
- Converts to Apache 2.0 on January 1, 2030
See CONTRIBUTING.md.
Built with Python, FastAPI, Qdrant, Alpine.js, and a lot of LLMs.