NetriqAI

Engineering intelligence for infrastructure teams

Infrastructure answers with the evidence attached.

NetriqAI classifies the question, retrieves from authoritative sources, and shows you what it used — then labels anything that changes device state for verification before it reaches production.

Open the app See how an answer is built
Networking & security Cited sources Verify before production

Free to start. It has no route into your devices, and nothing you paste trains a model.

A replay of NetriqAI answering a question about an eBGP session flapping: it classifies the query as routing on Cisco IOS-XE, retrieves RFC 4271, Cisco IOS-XE and NIST sources, then returns two show commands with an eighty percent confidence label marked verify before production.

Analyze

Paste a configuration. Get the gaps, with citations.

Map a device configuration against the standard you are actually audited on, and get findings that name the control they came from.

  • PCI-DSS, NIST SP 800-53, CIS Benchmarks, ISO 27001, NIST CSF, or India's DPDP Act
  • Every finding cites its requirement number, so the report is defensible
  • The same workspace family covers config analysis, version diff and hardening scores

The run shown scored a deliberately weak configuration 20/100 — and reported its own confidence at 40%, rather than presenting the result as certainty.

A PCI-DSS compliance report scoring a deliberately weak router configuration at 20 out of 100, with compliant and non-compliant controls listed by requirement number.

How an answer is produced

Retrieval first, then reasoning.

A general-purpose model will answer a BGP question confidently whether or not it knows the answer. NetriqAI puts a retrieval and grounding stage in front of the model, and carries the result through to what you read.

  1. 01

    Classify the question

    Query type, vendor, and domain are resolved first, so retrieval is scoped to the right corner of the knowledge base rather than searching everything.

  2. 02

    Retrieve candidate sources

    Semantic vector search and keyword search run together, their results are merged and reranked, and metadata filters keep vendor-specific material with its vendor.

  3. 03

    Reason over what was retrieved

    The model answers with those sources in context. Where retrieval returns nothing relevant, the answer says so instead of inventing detail.

  4. 04

    Classify the output for safety

    Every answer carries a safety label and a confidence figure. Anything that changes device state is marked for verification before production, not presented as tested.

Threats

The vulnerabilities that are actually being exploited.

Threat Intel reads the CISA Known Exploited Vulnerabilities catalogue and recent critical CVEs from NVD — both public, both authoritative, with no third-party feed in between.

  • Filter by vendor to see only what touches your estate
  • Each entry states plainly what to do about it, with the remediation date
  • CVE Lookup covers severity, exploit status and references for a specific ID
Threat Intel listing CISA actively-exploited vulnerabilities with remediation guidance and due dates.

Utilities

The deterministic things stay deterministic.

Subnet maths is arithmetic, not inference. The IP and subnet tools run entirely in your browser and return the same answer every time.

  • Covered by their own tests — including /31 point-to-point links and /32 host routes
  • Non-contiguous masks are rejected rather than quietly miscalculated
  • Nothing leaves the browser, so there is nothing to send anywhere

Not everything should be answered by a model. Where an exact answer exists, NetriqAI computes it.

Subnet calculator showing network, broadcast, host range, masks and binary representation for 192.168.1.0/24.

Everything else it does today

Six places to work, not a drawer of tools.

Each workspace is organised by the job you are doing. Everything listed here is built, tested, and running today.

Ask

  • Chat

Plain-language questions across networking, security, and infrastructure, answered with sources.

Analyze

  • Config Analyzer
  • Config Diff
  • Security Score
  • Compliance

Paste a configuration to find security gaps, compare two versions, score hardening, or map against a standard.

Threats

  • CVE Lookup
  • Threat Intel

CVE detail from NVD, and actively exploited vulnerabilities from the CISA KEV catalogue.

Change

  • Config Migration
  • CLI Translator
  • Runbook

Convert between vendor platforms and draft procedures. Output is a draft to review, never a tested change.

Diagnostics

  • Syslog
  • Packet Analyzer
  • Topology

Paste syslog or tcpdump output for analysis, and draw a network map you maintain yourself.

Utilities

  • IP & Subnet
  • Config Templates

Subnet maths and address lookup that run entirely in your browser, plus local template storage.

Security and privacy

It reads what you paste. Nothing else.

No training on your data

Your queries and configurations are not used to train models. Chat history is encrypted, and account deletion removes your data including API keys and usage records.

No access to your devices

NetriqAI has no path to your infrastructure. It cannot log into a device, push a configuration, or scan a subnet. Everything it knows about your network, you told it.

Answers you can audit

Answers record the sources retrieved and the reasoning path taken, so a surprising result can be traced rather than argued with.

Regulatory posture

Built to India's DPDP Act 2023 and GDPR expectations, with a documented retention period and a deletion path that states exactly what is removed.

Generated change is treated as a draft

Migrations, translated commands and runbooks are labelled for verification. Nothing produced by a model is described as tested on hardware, because it has not been.

Built on Cloudflare

Runs on Cloudflare Workers with data in D1 and Vectorize, served from the edge, with request-level observability.

What is ready, and what is not

We would rather tell you where the edges are.

Plenty of products describe a roadmap in the present tense. This is where NetriqAI actually stands.

Live

The fifteen workspaces above, and the public API

Built, tested, and in use. A versioned /v1 API covers chat, config analysis, security scoring, runbooks, CLI translation, CVE lookup and threat intelligence, with per-key rate limits and usage accounting.

Deepening

Knowledge depth per domain

Routing, firewalls, VPN and zero-trust material is grounded in authoritative public sources such as NIST and RFCs. Coverage is stronger in some areas than others, and we do not claim comprehensive support for a vendor until its retrieval and answer quality has been measured.

Planned

Voice and collaboration depth, MCP, and a CLI

Deep coverage for Cisco Webex Calling, Control Hub, CUCM and CUBE is the next domain pilot. A Model Context Protocol server and a terminal client follow full API certification. None of these are available yet.

Planned

Connecting to your own network

A read-only site connector, and later any form of assisted remediation, require explicit customer authorisation and a human approving every change. Today NetriqAI reads only what you paste into it.

Try it against something you already know

Ask it something you already know the answer to.

That is the honest way to evaluate a tool like this. Bring a question you have already solved, and see whether the answer holds up — sources, caveats and all.

Open the app Talk to us