Skip to content

ParendumOU/Nexora

Nexora

Nexora

Multi-tenant AI-agent orchestration platform. Define agents with personas, skills, and tools. Let them chat in real time, execute tools, decompose work, delegate to sub-agents, and stream results β€” all self-hosted, on your own infra.

License Release Last commit PRs welcome Stars

Python FastAPI Next.js PostgreSQL Docker

Nexora β€” AI-agent orchestration platform (click to watch the full demo)

🌐 Website Β· πŸ“– Docs Β· 🧩 Marketplace


What is Nexora?

Nexora is the free, MIT-licensed OSS core of an AI-agent orchestration platform. Users build agents (personas + skills + tools), and those agents collaborate in real time: they run tools, break work into tasks, spawn bounded sub-agents, and stream their output over WebSocket or SSE.

This repository is pure platform β€” zero billing, licensing, or paywall logic. The paid self-hosted product (NexoraCloud) consumes this repo and layers commercial features on top.

Why Nexora?

  • πŸ”’ Self-hosted, your data stays yours. Runs entirely on your own infra β€” no agent traffic leaves your network.
  • πŸ”Œ No vendor lock-in. ~46 LLM providers behind one interface; swap Claude ↔ GPT ↔ Gemini ↔ local Ollama without touching your agents.
  • 🀝 Real multi-agent orchestration. Agents delegate to sub-agents, run tools, and stream results in parallel β€” not just a single chat loop.
  • 🧩 Extensible by design. Knowledge bases (RAG), semantic memory, ~90 built-in tools, and a package marketplace.
  • πŸ†“ MIT-licensed core. Use it, fork it, ship it. Commercial features are opt-in via NexoraCloud.

Use cases

  • Internal AI ops assistants that touch Slack, Jira, Kubernetes, and your own APIs.
  • Research / RAG agents grounded in your private knowledge bases.
  • Automation crews that decompose a goal into tasks and execute them across sub-agents.
  • A self-hosted alternative to closed agent platforms, with full control over models and data.

Highlights

  • Agent builder β€” visual React-Flow graph: personas, skills, tools, sub-agents, bounded delegation.
  • ~46 LLM providers (+3 OAuth) β€” Claude, Gemini, OpenAI, Ollama, Vertex, Bedrock, Azure, Cohere, DeepSeek, Groq, Mistral, xAI, Perplexity, Together, Fireworks, OpenRouter, and more.
  • ~90 built-in tools / ~15 skills β€” Slack, Discord, Jira, Linear, Notion, PagerDuty, Google Drive, S3, Kubernetes, Playwright, hardened http_request (SSRF allowlist), agent-to-agent messaging.
  • Knowledge base / RAG (pgvector) + semantic memory search + multimodal image input.
  • Real-time streaming over WebSocket, with an SSE alternative.
  • Multi-tenant auth β€” email/password (Argon2), JWT, API keys, invite-only mode, TOTP 2FA.
  • Marketplace client β€” install community skills/tools/personas/agents with dependency auto-install.
  • Recovery engine β€” retries, circuit breaker, stale-heartbeat watchdog.
  • Full-platform backup / restore β€” export a whole instance (or one org) to a portable ZIP.
  • Clients β€” web UI, terminal client (NexoraCLI), and a mobile app (NexoraMobile).

Screenshots

Chat Orchestration (sub-agents + task tree)
Nexora chat Nexora orchestration

More in the documentation.


Tech stack

Layer Tech
Backend Python 3.12, FastAPI, SQLAlchemy 2 (async), Alembic
Frontend Next.js 15 (App Router), TypeScript, Tailwind, Zustand, React Flow
Data PostgreSQL 16 + pgvector, Redis 7
Proxy nginx 1.27
Runtime Docker Compose

Quick start

git clone https://github.com/ParendumOU/Nexora.git
cd Nexora

cp .env.example .env          # then set SECRET_KEY + ENCRYPTION_KEY (see SETUP.md)

make dev                      # dev stack: backend :8000, frontend :3000, nginx :8080
# or
make up                       # production stack (nginx on HTTP_PORT, default 80)

docker compose exec backend alembic upgrade head   # run migrations on first boot

First visit with no users β†’ /setup to create the admin account. Full instructions in SETUP.md.

Common commands

make dev      # dev (hot reload)
make up       # production
make down     # stop
make logs     # tail logs
make clean    # stop + wipe volumes (DESTRUCTIVE)

Project layout

backend/        FastAPI app (agents, orchestration, providers, tools, RAG, auth)
frontend/       Next.js 15 web UI
nginx/          reverse proxy config
docker-compose*.yml   prod / dev / data-only stacks
SETUP.md        standalone setup guide

Clients

Client Repo
Terminal (Go TUI) ParendumOU/Nexora-CLI
iOS / Android ParendumOU/Nexora-Mobile

Contributing

Issues and PRs welcome. See CONTRIBUTING.md and our CODE_OF_CONDUCT.md. For security reports, see SECURITY.md.


License

MIT © Parendum OÜ


⭐ Like what you see?

If Nexora saves you from wiring agents together by hand, drop a star β€” it helps others discover the project and directly shapes the roadmap. Got a question or want to show what you built? Open an issue or join us at nexora.parendum.com.


Star history

Star History Chart

About

Self-hosted, MIT-licensed AI-agent orchestration platform. Agents with personas, skills & tools that chat, run ~90 tools, delegate to sub-agents, and stream results across ~46 LLM providers. RAG, semantic memory, marketplace.

Topics

Resources

License

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Contributors