Skip to main content
Admin42

Product

What NELDA does.

NELDA — Network Engineer and Local Device Administrator

NELDA is a virtual network admin appliance. She plugs into the network, learns the environment, and helps with everything a network admin would do — in plain language, by conversation, on your premises.

Human-In-The-Loop by design. NELDA recommends. You approve. She executes. No autonomous action against your environment — ever.

nelda.local/admin

new conversation

local backend

⌘K
What is the Capital of the US?

local 0.26s · 31 tokens

I don't know — that's outside my area. If you have a larger backend configured, you might want to escalate that question to it.
ask anything · drop a config to upload · ⌘↵ to send

Preview — representative interface

What NELDA is

A network administrator who happens to use AI.

NELDA is purpose-built for network administration. She runs on your premises, watches your gear, and — when you approve — executes the configuration changes she recommends. Narrow by design, accurate where it counts.

What NELDA isn't

A general-purpose AI or a search engine.

NELDA won't browse the web for you, write your essays, or answer trivia. Ask her the capital of the US and she'll tell you she doesn't know — and offer to route the question to a larger backend if you've configured one. That's the point.

Capabilities

What NELDA does on day one.

  • 01 · Setup flow

    She introduces herself.

    On first boot, NELDA reports her IP, gateway, and DNS — then walks you through pinning her address, configuring DNS & NTP, authentication (LDAP / AD or TACACS+ / RADIUS), and VM provisioning. Skip anything you're not ready for; she'll come back to it later.

  • 02 · Ask, integrate, or build

    She fits the tools you already have.

    For every supporting service — NMS, DHCP, DNS, IPAM, configuration management, syslog, ticketing — NELDA first asks what you already run, integrates with it if you have one, and offers to build one if you don't. Your existing stack doesn't get replaced. It gets a teammate.

  • 03 · Discovery & monitoring

    She finds your network.

    Passive discovery on power-up — ARP, DHCP leases, LLDP / CDP — with active scans (SNMP, ping sweeps, targeted probes) only after you approve them. She maps devices, subnets, and existing tools, and surfaces what's actually worth your attention.

  • 04 · Recommend, approve, execute

    Human-In-The-Loop, every time.

    NELDA describes what she's seeing in plain language, proposes a fix, and shows the exact change she would make. You approve, defer, or skip. When you approve, she executes the change. When you defer, she waits. Nothing touches your environment without your go-ahead.

  • 05 · Honest uncertainty

    She says "I don't know."

    NELDA is narrow by design — and built to detect her own uncertainty and surface it rather than guess. When she doesn't know, she tells you and offers to route the question to a bigger backend if one is configured. That same uncertainty signal is part of how NELDA decides whether to escalate — see how NELDA thinks below.

  • 06 · Security advisories

    She watches for trouble that's actually yours.

    Continuous cross-referencing of CISA KEV, NVD / MITRE CVE, MITRE ATT&CK, and vendor PSIRTs against your discovered inventory. NELDA tells you which advisories apply to gear you actually run — patch path, mitigation, or workaround — never generic alerts.

  • 07 · Scales with you

    One interface, more horsepower.

    NELDA runs locally out of the box. When you need more capacity or specialized skills, expand on your terms — another GPU, a dedicated GPU server, an all-in-one box like an NVIDIA DGX Spark or AMD Strix Halo, or an external LLM. We favor the all-in-one path for its simplicity, but the choice is yours.

  • 08 · Curated, fine-tuned models

    The right model for the job.

    We've selected the best models for each kind of task and fine-tuned them for network administration. NELDA also uses retrieval-augmented generation over publicly available documentation from the major networking vendors. When you have authorized access to official vendor documentation, NELDA can be further trained on your materials for environment-specific accuracy.

  • 09 · On-premise & private

    Your data stays yours.

    NELDA runs on your network. Telemetry, configurations, and logs do not leave your environment unless you explicitly choose to route a question to an external backend. Privacy is structural, not promised.

How NELDA thinks

Four levels. One conversation. The human is always last.

NELDA escalates only as needed. Routine work stays on the appliance; harder questions reach for bigger models; every path that touches your environment ends with you.

  1. Level 1 · Always present

    Onboard LLM

    A small, fine-tuned model running on the NELDA appliance itself. Handles routine Q&A, basic configuration suggestions, and most day-to-day work locally.

  2. Level 2 · Optional

    Specialized LLM on your hardware

    Larger or vendor-specialized models running on your own hardware — DGX Spark, Strix Halo, a GPU server you build. NELDA orchestrates loading, routing, and use.

  3. Level 3 · Optional

    External LLM

    Cloud-hosted models reached over the API — Gemini, ChatGPT, Claude — when extra capacity or capability is worth the round-trip. Off by default; on only if you turn it on.

  4. Level 4 · Always

    The human

    You are the final authority on anything that touches your environment. Every recommendation from L1, L2, or L3 ends here for your approval. That's Human-In-The-Loop.

Common escalation paths

NELDA picks the path that fits the task.

  • Routine: L1 → L4 (onboard answers, you approve)
  • Specialized: L1 → L2 → L4 (onboard hands off to your bigger model, you approve)
  • External help: L1 → L3 → L4 (onboard reaches an external LLM, you approve)
  • Full chain: L1 → L2 → L3 → L4 (when the question warrants every tool)

Want to see NELDA in your environment?

We work with a small number of early customers at a time. If your network sounds like the kind of place NELDA was built for, we'd like to hear about it.