Ask Your Network Anything With the Pico5G Copilot
The Pico5G’s on-box AI test engineer drives the unit’s own tools, reads live signaling, searches 8,000+ documentation sections and answers with citations you can check. Diagnosis with receipts on a complete LTE/5G/NTN network in one box.
100%
Root-cause exam score with quality gate ≥80%
Every Conclusion Cited
Tool result, trace line, or document section
Approval-Gated
Engine-enforced safety outranks even your click
8,000+ Sections
On-box knowledge, 3,400+ figures — air-gap capable
What the Pico5G Copilot Does Today
Diagnose, With Receipts
Ask in plain language and the Pico5G Copilot drives the box’s own tools, checks a curated fault library built from real bench incidents, diffs your failing against golden reference traces, and searches the on-box corpus — Amarisoft Tech Academy and the ShareTechnote handbook — figures included. Expand the evidence trace under any answer and check its work.
Compose and Run Tests
“Sweep my modem across its Cat-M1 bands” becomes a composed, parameterized, executed and interpreted test campaign, the box even cycles the device’s radio automatically, and finished with a branded evidence report.
Watch the Bench
Every night, if the bench is idle, the sentinel runs a baseline test and flags drift — deterministically, no AI model in the loop. Bench health stops being a matter of opinion and becomes a log file.
Approval-Gated, Safety-Bounded, And Audit-Logged
The Pico5G Copilot diagnoses read-only. To change anything, such as retune a cell, cycle a modem, or start a test, it stops and asks, naming the exact action and arguments. Your approval isn’t the last word: the unit’s engine-enforced safety policy (TX-gain caps, band allowlists) is checked after you click, so no one — human or AI — talks the box past its limits.
Every state-changing action lands in an append-only audit log with its actor. Prefer advise-only? One policy setting removes the AI’s action tools entirely.
Tested Like Everything Else On the Bench
The Pico5G ships an accuracy benchmark that seeds real faults on the real hardware, asks the copilot the operator’s question, and scores it against the known cause — including a healthy-control case to prove it doesn’t invent problems. The root-cause exam passed at 100%; the standing quality gate is ≥80%, and no learned knowledge is promoted unless the benchmark still passes. Use the same exam to vet any model — cloud, LAN, or on-box — before you trust it.
What It Can't Do — On Purpose
The Pico5G Copilot cannot act without you. It doesn’t know your device’s proprietary internals. It will not run unattended overnight experiments without authorization. The only thing that runs by itself is the deterministic sentinel, configured by you. And it cannot exceed the radio safety policy, even with your approval.
Model-Agnostic And Air-Gap Capable
Bring an Anthropic, OpenAI, or GLM key. The AI model is stored on the unit, never displayed back, never sent to the browser, and never seen by support. Or bring no key at all: point the Pico5G Copilot at any OpenAI-compatible model on your LAN or on the box itself. The knowledge corpus, fault library, and golden traces live on the unit and work air-gapped.
44 Tools, Same Rules
The MCP server exposes the same 44 typed tools the GUI and CLI use to any MCP-capable client — Claude Code, Claude Desktop, or your own agent — bounded by the same safety policy, license gating, and audit log.
Where the Pico5G Copilot is Going Next
In Validation
Investigator Mode
Multi-layer root-cause hunts that capture live PHY-to-NAS traces around an event and run bidirectional A/B config experiments to turn a plausible cause into flip-verified evidence.
On the Roadmap
Autonomous Investigator Mode
As a product capability, multi-unit fleet scenarios, and a fully air-gapped on-box model option for secure facilities.
Related Content
All the functionality and benefits of the next leading cellular testing platform in a portable compact chassis. Offered in multiple models.
The PicoNet accelerates enterprise and industrial private network deployments with simplicity, flexibility, and carrier-grade performance.
A satellite-link test bench built into the Pico5G box you already own. Fly real LEO, MEO and GEO passes over a live 5G NR or NB-IoT cell.
Learn more about the specifications and functionality of the Pico5G Series of products.
Frequently Asked Questions
For a cloud model, yes — you bring your own Anthropic, OpenAI, or GLM key. You enter it yourself; it is stored on the unit, never displayed back, never written to the browser, and never visible to Nutaq support. Alternatively, bring no key: point the copilot at any OpenAI-compatible model server on your LAN or on the box.
Your question and the tool results the diagnosis gathers — configuration, signaling excerpts, KPIs: bench test data, no personal data. With a local/LAN model, nothing leaves the unit.
Two mechanisms: citations — every answer carries an expandable evidence trace with the exact tool calls and sources — and the benchmark, which seeds real faults on the real hardware and scores the copilot against the known cause. The root-cause exam passed at 100%; the standing gate is ≥80%.
Honest answer: it depends on the model you bring — and the built-in benchmark exists exactly for this. Vet a candidate local model against the same seeded-fault exam before trusting it.
No. Any change requires your explicit approval, and the engine-enforced safety policy is checked after the approval. The only thing that runs unattended is the deterministic nightly sentinel, which you configure.