network-aiops

Disclaimer: This is a community-maintained open-source project and is not affiliated with, endorsed by, or sponsored by Cisco, Arista, Juniper, NetBox Labs, or any network vendor. Vendor and product names are trademarks of their respective owners. Source code is publicly auditable at github.com/AIops-tools/Network-AIops under the MIT license.

Governed multi-vendor network device operations for AI agents — 32 MCP tools, every one wrapped with the bundled @governed_tool harness: a local unified audit log under ~/.network-aiops/, policy engine, token/runaway budget guard, undo-token recording, and graduated-autonomy risk tiers. Credentials (device passwords + the NetBox token) are kept in an encrypted store (secrets.enc), never plaintext on disk.

Devices are reached over NAPALM; an optional NetBox block adds source-of-truth lookups.

Standalone: the governance harness is bundled in the package (network_aiops.governance) — network-aiops has no external skill-family dependency. Coverage focuses on common device operations and is not yet exhaustive.

Verification status: the test suite is mock-based; not yet validated against live devices — self-testable with cEOS / vMX / containerlab. See docs/VERIFICATION.md.

What works

Read device facts, interfaces (+ counters/IP), BGP/LLDP neighbors (summary and detail), ARP/MAC tables, VLANs, route lookups, hardware environment, optics, NTP, users, SNMP info, VRFs, and an aggregated device_health; run read-only RCA diagnostics that flag down/erroring/flapping interfaces and unhealthy BGP neighbors — each finding citing the measured number that tripped it; back up the running config, dry-run a config diff, and merge/replace/rollback config — across the five core NAPALM platforms below. Optional NetBox lookups (devices + interfaces) confirm intended state before a change.

NAPALM does not implement every getter on every platform; an unsupported getter returns a teaching error ("not supported by the <driver> driver") rather than crashing. Secrets are never returned — get_users redacts password hashes and get_snmp_information redacts community strings.

Supported devices

Additional platforms (Nokia SR OS / SR Linux, Huawei VRP, etc.) are reachable via NAPALM community drivers but are not officially tested here. Need one? See Contributing.

Supported actions

Security: read-only mode

This tool is meant to be handed to an AI agent, so its safety story is enforced by the server rather than requested in a prompt:

export NETWORK_READ_ONLY=1

With that set, the 4 write tools are never registered. An MCP client lists 28 tools instead of 32 — the writes are not hidden, not gated behind a flag, and not merely refused when called. They are absent from the session. A model cannot invoke a tool it was never offered, and cannot be argued into one.

That distinction is the whole point. A tool that exists but refuses still invites retry loops and "I'll describe the call instead" behaviour from smaller models, and it leaves a reviewer trusting a promise. An absent tool is a fact you can check: connect, list the tools, and see that the writes are not there.

Enforcement is two layers deep, so the switch cannot be sidestepped by changing entry point:

Read operations are unaffected, and every call is still audited to ~/.network-aiops/audit.db.

The read/write split is derived from each tool's declared risk_level, and a test asserts that this never disagrees with the [READ]/[WRITE] tag in the tool's own documentation — so a write can't quietly present itself as a read.

Running a smaller / local model? See agent-guardrails.md — it lists the guardrails this tool now enforces for you (so you don't spend prompt budget restating them) and gives a ready-made system prompt for what's left.

Quick Start

uv tool install network-aiops
network-aiops init                                  # wizard: device + driver + host + encrypted password
network-aiops doctor
network-aiops device facts -t core-sw1
network-aiops device health -t core-sw1
network-aiops diagnose interface-health -t core-sw1   # worst-first interface RCA
network-aiops diagnose bgp -t core-sw1                # worst-first BGP-neighbor RCA
network-aiops config backup -t core-sw1 -o core-sw1.cfg

Playbook: triage a flaky uplink before touching config

# 1. Ask the device what's actually wrong — findings are ranked worst-first and
#    each cites the measured value (error count, last-flap seconds, uptime).
network-aiops diagnose interface-health -t core-sw1
network-aiops diagnose bgp -t core-sw1

# 2. If interface-health flags a link admin-up/oper-down with climbing errors and
#    BGP shows the peer on that path recently reset, you have your root cause: a
#    physical-layer fault (cable/optic) resetting the session — not routing.

# 3. Confirm intended state, then remediate with the governed, audited path.
network-aiops device counters -t core-sw1
network-aiops config diff -t core-sw1 -f fix.cfg      # dry-run the change first

Create ~/.network-aiops/config.yaml:

devices:
  - name: core-sw1            # used as -t core-sw1
    driver: eos               # ios | nxos | nxos_ssh | iosxr | eos | junos
    host: 10.0.0.1
    username: admin
    optional_args:            # passed verbatim to NAPALM (optional)
      secret: enable-pw       # enable/secret
      port: 443
# Optional source-of-truth:
netbox:
  url: https://netbox.example.com

Secrets are stored encrypted in ~/.network-aiops/secrets.enc (Fernet/AES + scrypt-derived key; chmod 600) — never in config.yaml or a plaintext .env. Device passwords are keyed by device name; the NetBox token uses the reserved name netbox-token:

network-aiops init                     # interactive wizard (recommended)
network-aiops secret set core-sw1      # store a device password (hidden prompt)
network-aiops secret set netbox-token  # store the NetBox API token
network-aiops secret list              # names only — values are never printed
network-aiops secret migrate           # import a legacy plaintext .env, then delete it

Export NETWORK_AIOPS_MASTER_PASSWORD to unlock the store non-interactively (MCP server / cron). Legacy plaintext env vars (NETWORK_<TARGET_UPPER>_PASSWORD, NETWORK_NETBOX_TOKEN) remain a deprecated fallback. An empty device password is allowed for key-based SSH auth.

MCP

{
  "command": "network-aiops",
  "args": ["mcp"],
  "env": {
    "NETWORK_AIOPS_CONFIG": "~/.network-aiops/config.yaml",
    "NETWORK_AIOPS_MASTER_PASSWORD": "…"   // unlocks the encrypted secret store
  }
}

Audit & Safety

• Every tool call is logged to ~/.network-aiops/audit.db (local SQLite; relocate with NETWORK_AIOPS_HOME).
• config_merge / config_replace capture the pre-change running config and record an inverse config_replace-to-backup undo descriptor.
• config_replace is risk_level=high; CLI destructive commands (config merge/replace/rollback) require double confirmation and support --dry-run (which prints the diff without committing).
• All device text passes through sanitize() (output hygiene: control/format-char stripping + truncation).
• Device passwords and the NetBox token live only in the encrypted secrets.enc (chmod 600); tools never return passwords, SNMP community strings, or hashes.

See skills/network-aiops/SKILL.md and SECURITY.md for details.

Companion Skills

Contributing & feature requests

Coverage is intentionally focused. Need a device or action that isn't here yet? Open an issue or pull request at github.com/AIops-tools/Network-AIops — contributions, feature requests, and comments are all welcome.

License

MIT — github.com/AIops-tools/Network-AIops