AIMusubi is a local-first agentic automation framework that brings LLM-driven intent-based operations to real network devices — Cisco IOS-XE, Arista EOS, and VyOS — through a unified API, reproducible bootstraps, and full-stack observability.
This is a framework that becomes a platform when bootstrapped.
If you want to see an LLM observe, decide, execute, and verify changes on actual routers in your own lab — AIMusubi is built for you.
- Local-first – Runs entirely in your lab. Your devices, your data, your API.
- Real devices – Talks to Cisco, Arista, and VyOS using real APIs.
- Unified intent engine – Same operations across all vendors.
- Agentic by design – Built specifically for LLM tool-calling workflows.
- Reproducible bootstraps – One script builds the full environment.
- Open-core – Clean separation between lab framework and future enterprise tier.
AIMusubi installs an entire Level-5 agentic NetOps stack:
- AIMusubi API (FastAPI)
/intent/exec,/openapi.json,/metrics,/health
- Vendor Adapters
- Cisco RESTCONF (YANG)
- Arista eAPI (JSON-RPC)
- VyOS REST/RESTCONF
- Intent Engine
- Vendor-agnostic operations (
iface.list,routing.v4.rib,ospf.neigh, etc.)
- Vendor-agnostic operations (
- SQLite Memory
- Credentials, observations, evolving state
- Observability
- Prometheus metrics + Grafana dashboards
- LLM Frontend
- Open WebUI wired directly to AIMusubi’s OpenAPI schema
AIMusubi includes a full operator-grade toolchain (nmap, masscan, fping, SNMP utilities, DNS tools, traceroute, iproute2, etc.) so your environment has the same diagnostic visibility as a real NetOps workflow. Please use in a lab environment only
All installed and wired together via bare-metal or Docker bootstrap.
Full installation details live in
docs/, but here’s the shortest path.
git clone https://github.com/aimusubi/aimusubi.git
cd aimusubiBare-metal (Ubuntu 22.04 / 24.04)
chmod +x bootstrap/aimusubi_l5_fullstack_baremetal.sh
sudo ./bootstrap/aimusubi_l5_fullstack_baremetal.shDocker stack
chmod +x bootstrap/aimusubi_l5_fullstack_docker.sh
./bootstrap/aimusubi_l5_fullstack_docker.shFollow:
docs/post_bootstrap_activation.mddocs/openwebui_setup.md
Then open Open WebUI, select your LLM, and AIMusubi is ready for tool-calling.
- Network engineers who want to experiment with LLM-driven NetOps
- Homelab builders running multi-vendor topologies
- SRE / DevOps engineers curious about agentic workflows
- Educators & students learning real infrastructure automation
- IT Operators Exploring the forward understanding of natural language operations
AIMusubi is a lab-first framework, not a production change control system.
- Version: 1.0.0 (Open-Core Lab Release)
- Vendors: Cisco IOS-XE, Arista EOS, VyOS
- OS Target: Ubuntu (bare-metal) + Docker stack
- Security Model: Local lab mode, self-signed certs accepted, http used for analysis
See:
docs/roadmap.mddocs/ARCHITECTURE.md
Start here:
docs/overview.mddocs/installation_baremetal.mddocs/installation_docker.mddocs/post_bootstrap_activation.mddocs/openwebui_setup.mddocs/adapters.mddocs/intents_reference.md
👉 Join the MusubiAG Discord Community
https://discord.gg/xAeXxM5f
- YouTube – The Agentic Engineer
- GitHub Issues – bugs, ideas, suggestions
- Roadmap –
docs/roadmap.md
Contributions are welcome — adapters, intents, dashboards, docs, improvements.
See:
CONTRIBUTING.mddocs/CHANGELOG.md
If AIMusubi helps you build, learn, or think differently about NetOps, that’s the mission.
AIMusubi communicates with network operating systems exclusively through standards-based or publicly documented APIs. No proprietary software, firmware, or intellectual property from any vendor is included in this repository.
Users must supply their own properly licensed network operating system images in accordance with the terms of their vendor agreements. AIMusubi does not provide, store, transmit, or facilitate access to any vendor images.
All vendor names and trademarks remain the exclusive property of their respective owners. AIMusubi is an independent open-source project and has no affiliation with Cisco, Arista, VyOS, or any other vendor referenced in examples or documentation.