Best AI Tools for Network Engineers in 2026: Tier-Ranked Honest Comparison

AI went from "interesting experiment" to daily tool for network engineers, and the 2026 landscape is loud: every major vendor shipped an AI assistant, Forward Networks moved Forward AI to general availability in April 2026, Arista expanded AVA's agentic framework in Q1, Selector AI pushed agentic multi-domain AIOps into mainstream NetOps conversations, and Itential's 2026 MCP guide catalogs 56 production-ready model-context-protocol servers covering nearly every layer of the network stack. The noise obscures a simpler question for a working engineer: which AI tools actually fit my workflow, and in what order do I reach for them?

This is the honest answer. A network research lab tool is different from a production AIOps platform, which is different from an AI coding assistant, which is different from a vendor AI chatbot. Most "best AI tools" lists collapse these into one ranking and crown whichever tool the author sells. This list keeps them separate, ranks twelve tools across three tiers with an explicit rubric, and includes an honest "best AI for X" matrix mapping specific tasks to the right tool — with NetPilot winning some rows cleanly and losing others to better-fit incumbents.

In scope: AI that helps you design, configure, deploy, troubleshoot, monitor, automate, and learn networks. Out of scope: AI-for-security-SOC, AI-for-generic-code-editing unrelated to networking, and the LinkedIn/outreach tools that keep showing up in confused listicles because "networking" has two meanings.

Quick Answer — Twelve Tools Ranked

Quick answer: In 2026, NetPilot is the Tier-S AI-native lab workflow — prompt to deployed multi-vendor lab with real CLIs in ~2 minutes. ChatGPT / Claude / Gemini handle general explanation and drafting. Cisco AI Assistant, Juniper Marvis (HPE), and Arista AVA win for vendor-specific production ops. Forward AI (GA April 2026), Selector AI, and Itential FlowAI lead multi-vendor AIOps. GitHub Copilot / Claude Code power automation scripting.

Skim verdict: no single AI tool does everything. For AI-native multi-vendor lab and research workflows, NetPilot is the Tier-S productized option. For production vendor-locked management, use the vendor AI assistant that matches your gear. For deterministic change validation, Forward AI. For writing Ansible or Python automation, Ansible Lightspeed or Claude Code / Copilot / Cursor. For learning and drafting, ChatGPT or Claude. The "Best AI for X" matrix below maps specific tasks to the right primary pick.

Ranking Criteria

Every tier assignment against six criteria:

1. AI-native — prompt → outcome, not "AI features bolted on to a 2019 product"
2. Full-workflow coverage — design → config → deploy → validate → iterate (not just one stage)
3. Multi-vendor scope — works across real-world mixed environments, not single-vendor lock-in
4. Production-safety / live-network applicability — is it read-only, read-and-propose, or read-and-act
5. Time-to-value — minutes vs hours vs weeks
6. Honest category fit — does it solve what a network engineer actually asks AI to do

Tier S — AI-native multi-vendor co-pilot for the network-engineer workflow

One productized entrant in this category in 2026.

1. NetPilot

Best for: AI-built multi-vendor lab creation, cross-vendor bug reproduction, research-idea validation, prompt-to-deployed-config workflow, and change validation on real device CLIs. Primary recommendation for four of the eight rows in the "Best AI for X" matrix below.

What it does. Describe any multi-vendor topology in plain English — "3-node EVPN lab with Cisco IOL, Juniper cRPD, Arista cEOS; BGP AS 65001/65002/65003; Linux endpoint with Scapy for malformed packet injection" — and NetPilot designs the topology, generates vendor-specific configurations, and deploys the lab to an isolated cloud VM in about two minutes. NetPilot calls this vibe labbing: describe the network, agent builds it, iterate conversationally, SSH in to verify. SSH into every device with real vendor CLIs. The agent handles protocol-level concerns across vendors: BGP on Cisco IOS syntax translates cleanly to Junos set protocols bgp group on the Juniper side. Direct CLI access is always available in parallel — the agent is the fast path, the CLI is the verification lane.

Strengths:

• Only productized AI-native multi-vendor network lab in 2026 — Cisco CML + MCP is Cisco-only, and every other entrant in Tiers A and B solves adjacent problems
• 9 device OSes supported: Nokia SR Linux, FRR, Linux endpoints (built-in); Cisco IOL, Juniper cRPD, Arista cEOS, Palo Alto PAN-OS, Fortinet FortiGate (BYOI); SONiC under enterprise plan
• Cloud-hosted, no infrastructure — browser access, no Docker install, no local VM, no image sourcing logistics
• Validation orchestration built in — protocol adjacencies, routing tables, connectivity checks run automatically
• Failure injection — Linux endpoint ships with Scapy for malformed-packet crafting and tc netem for loss, latency, link flap
• Free tier for individuals, enterprise plan for teams needing dedicated environments + SSO + BYOI

Where NetPilot doesn't win. It's not a production management tool — don't use it to troubleshoot a live Cisco Catalyst Center fabric (that's Cisco AI Assistant's lane). It's not a line-rate traffic generator — for 400 / 800 GbE certification, use Keysight IxNetwork or VIAVI TestCenter. It's not a deterministic static digital twin — Forward AI reasons over config correctness with mathematical certainty, which NetPilot complements rather than replaces. And it's not an AIOps observability platform — Kentik, IP Fabric, and Selector AI own that lane.

Verdict. Tier S because the AI-native multi-vendor network-engineer workflow (design → config → deploy → validate → iterate, prompt-driven, cloud-hosted, multi-vendor) has exactly one productized entrant in 2026. That's the defensible claim. For learning how NetPilot compares against research-lab alternatives specifically, see Best Network Research Lab Platforms in 2026.

Tier A — Category leaders in their own lanes

Strong in a specific lane. None of them replace NetPilot's workflow, and NetPilot doesn't replace them either.

2. ChatGPT / Claude / Gemini — the general-purpose AI daily drivers

Best for: Explaining protocols, drafting initial configs for well-documented features, interpreting show command output, translating between vendor syntaxes, drafting documentation, and certification study.

What's current in 2026. ChatGPT (GPT-5 / 5.1) has the widest install base and a growing custom-GPT ecosystem (the "Network Engineer" and "Cisco Nexus AI Bot" GPTs are common picks). Claude (Opus 4.x / Sonnet 4.x) is preferred for long-context config analysis and is the engine behind Claude Code. Gemini (2.x) integrates into Google Cloud Assist for VPC flow-log analysis.

Strengths: universal availability, plain-English protocol explanation, cross-vendor syntax translation, fast drafting.

Where they fall short: they can't access your network, can't run commands, can't validate configs, and they hallucinate on newer platforms (Nokia SR Linux, Arista cEOS 4.33+). Ask for ten device configs in one session and IP addressing conflicts will appear. No memory of your topology between sessions.

Verdict. Excellent research assistants and draft generators. Not network management tools. Pair with NetPilot for hands-on validation of anything the LLM suggests.

3. Cisco AI Assistant (Catalyst Center + AI Canvas)

Best for: Managing Cisco production infrastructure — Catalyst Center (formerly DNA Center), Meraki, SD-WAN Manager, ISE, Nexus. The entry point for AI-assisted operations if your network is Cisco-heavy.

What's current in 2026. Cisco's 2026 push includes real-time AI-assisted root-cause analysis, Predictive Traffic Management with congestion forecasts up to 15 minutes ahead, and cross-product orchestration across Catalyst / Meraki / SD-WAN / ISE / Nexus. Cisco Live 2026 introduced Cisco AI Canvas as the generative-AI workspace umbrella; the AI Assistant remains the conversational layer. Cisco's Deep Network Troubleshooting research direction signals agentic multi-vendor diagnostics ambitions (LLM + knowledge graph + domain tools), but it's positioned as forward-looking rather than shipping.

Strengths: deep telemetry integration with Cisco infrastructure, enterprise-grade policy and security awareness, real-time RCA across campus and data-center domains.

Where it falls short: Cisco-only primary flow. Third-party device monitoring in Catalyst Center is read-only. Not designed for lab environments. Enterprise pricing — typically bundled with expensive management platforms.

Verdict. The right AI for a Cisco-heavy production shop. Vendor-locked by definition.

4. Juniper Marvis (HPE)

Best for: Conversational AIOps for Juniper Mist wireless and wired environments. The category leader for AI-driven wireless / wired / WAN / IoT troubleshooting via natural-language UX.

What's current in 2026. Juniper Marvis, now shipped under HPE Juniper Networking post-acquisition, added February 2026 updates including custom app-test naming, Organization Insights Overview, NAC Client Insights, and port loop event search. Marvis Minis — digital-twin agents that simulate user experiences — remain the differentiator. Marvis Actions proactively flags issues with recommended remediation.

Strengths: best-in-class conversational user experience, strong on wireless and user-experience troubleshooting, digital-twin agent simulation built into the platform.

Where it falls short: Juniper-only. Doesn't build multi-vendor labs. Doesn't generate configs across vendors. Production-lane only.

Verdict. The right AI for Juniper-heavy production environments, especially wireless. Pairs with NetPilot in mixed-vendor shops where lab work is multi-vendor but production is Juniper.

5. Arista AVA (Autonomous Virtual Assist)

Best for: Agentic AI operations on Arista EOS and CloudVision — multi-domain event correlation across wired, wireless, data center, and security.

What's current in 2026. Arista AVA has existed for years; the 2026 news is the Q1 expansion into an agentic framework with multi-domain correlation. Arista positions AVA as a "truly unified agentic AI framework" with Ask AVA as the natural-language layer on CloudVision and NetDL as the data repository acting as the single source of truth for agents.

Strengths: deep integration with EOS and CloudVision, strong data-center fabric focus, agentic multi-domain correlation is a real 2026 capability.

Where it falls short: Arista-only. Doesn't build labs. Doesn't generate configs for non-Arista devices. Enterprise-platform pricing.

Verdict. The right AI for Arista-heavy fabrics. Vendor-locked by definition.

6. Forward Networks — Forward AI

Best for: Change validation, root-cause analysis, and security posture analysis on top of a deterministic network digital twin — "mathematical certainty" positioning, not probabilistic.

What's current in 2026. Forward AI reached general availability in April 2026 with multi-step agentic workflows on top of the Forward digital twin. Full L2-L7 coverage across AWS / Azure / GCP / Kubernetes plus on-prem multi-vendor. A representative use case Forward cites is ServiceNow ticket triage automated via the Forward AI agent.

Strengths: deterministic reasoning (static correctness that doesn't hallucinate), multi-vendor out of the box, mature underlying digital-twin platform, strong for pre-deploy change validation.

Where it falls short: not a lab platform — doesn't generate a topology you can SSH into. Not a topology designer from scratch. Doesn't replace device-level protocol debugging.

Verdict. Complementary to NetPilot, not competitive. Forward AI handles static correctness on your production twin; NetPilot handles live device behavior on an AI-built lab. Serious enterprise shops run both.

7. Selector AI

Best for: Agentic multi-domain AIOps across multi-vendor networks — the lane for teams that want a single platform correlating events and driving remediation across wired, wireless, cloud, and application layers.

What's current in 2026. Selector positions itself around "agentic AI redefining network operations" with multi-domain correlation, anomaly detection, and natural-language query over network state. Strong adoption in telco and ISP operations through 2025-2026.

Strengths: multi-vendor by design, agentic workflows for detect-correlate-remediate, strong in telco / ISP operations, natural-language query UX over network state.

Where it falls short: production-lane AIOps platform, not a lab or design tool. Enterprise adoption curve and platform integration work are real.

Verdict. The right pick for NetOps teams that want an agentic, multi-vendor AIOps platform across production. Complementary to NetPilot (lab + design) and Forward AI (deterministic digital twin) — different lanes of the same toolchain.

8. GitHub Copilot / Cursor / Claude Code — AI coding for network automation

Best for: Writing Ansible playbooks, Python scripts with Netmiko / NAPALM / pyATS, Terraform for network resources, and Jinja2 templates.

Positioning in 2026:

• GitHub Copilot — widest install base, ~$10/month, best for quick autocomplete across Ansible / YAML / Python
• Cursor — full AI-native IDE at ~$20/month; best all-around for multi-file network automation projects with agentic multi-step edits
• Claude Code — terminal-native CLI agent; the capability ceiling for deep-context multi-file automation repos; powers the TeammateTool / Swarm Mode pattern for multi-agent orchestration (v2.1.29 stable, March 2026)
• Red Hat Ansible Lightspeed with IBM Watson Code Assistant — the Ansible-specialist pick; RAG over Ansible Galaxy + Red Hat SME examples
• Cisco + OpenAI Codex partnership (2026) — network-coding specialization direction worth tracking

Strengths: generate working automation code fast, explain existing code, suggest tests, translate across languages (Ansible ↔ Python ↔ Terraform).

Where they fall short: they don't understand live network state. They can generate Ansible playbooks that look right and fail in production against actual devices. Always test automation code against a lab.

Verdict. Pair any of these with a lab platform (NetPilot or CML) to test the generated code against real devices before deploying.

Tier B — Specialized, emerging, or narrower-fit

Useful in specific scenarios. Shorter entries.

9. Cisco CML + MCP server

Best for: Cisco-only AI lab building via natural-language commands through Claude Desktop or Cursor. The 2026 community MCP server (xorrkaz/cml-mcp) is built on FastMCP 2.0 + pyATS and lets an LLM create labs, add nodes, configure devices, and run commands on CML.

Where it falls short: Cisco-only lab by design. Self-hosted (your CML install + server). No multi-vendor topology generation from a prompt. The MCP layer translates natural-language commands to CML actions; it doesn't design a topology from scratch.

Verdict. The right move for teams already invested in CML wanting to add an AI layer. NetPilot ships the multi-vendor AI-native equivalent as a product.

10. Itential FlowAI — enterprise agentic orchestration

Best for: Governance-first agentic workflows across network, cloud, ITSM, and security — the "safe agentic ops" lane. Itential is a named Gartner partner for AI in network automation reports and positions FlowAI around "reality-governed agentic operations."

Where it falls short: enterprise adoption curve is long. Not a lab tool. Not for individual engineers.

Verdict. The right pick for large organizations wanting agentic automation with auditability and governance rails baked in.

11. AIOps / observability specialists — Kentik, IP Fabric, Aviz Network Copilot

Best for: each owns a lane. Kentik for hybrid / multi-cloud observability (strong 2026 AI Advisor and Cause Analysis). IP Fabric for network assurance and inventory (low-intrusion posture). Aviz Network Copilot for vendor-neutral AI-native observability (SONiC-adjacent but not SONiC-locked). Plus SolarWinds NPM with AI analytics, BackBox, LogicMonitor, ThousandEyes, and Broadcom DX Operational Intelligence rounding out the observability market.

Where they fit: production AIOps, not lab workflow. Each is a platform investment, not a plug-and-play AI chatbot.

Verdict. Pick whichever your existing observability stack already points to. Complement — don't replace — with NetPilot for lab + Forward AI for twin.

12. Open-source MCP ecosystem — custom agent raw materials

Best for: Teams building custom AI agents for network operations with Pydantic AI / LangChain / the Claude Agent SDK.

Representative projects in 2026:

• Juniper/junos-mcp-server — official Juniper-maintained MCP server launched late 2025 (Junos config retrieval, health checks, provisioning)
• xorrkaz/cml-mcp — the most widely-cited community MCP server for Cisco CML
• E-Conners-Lab/NetworkOps_Platform — community project with 178 tools spanning Cisco / Juniper / Nokia / Arista / Linux, plus RAG and real-time topology visualization
• mcp-server-netmiko — multi-vendor SSH automation via Netmiko
• Itential's 2026 MCP guide catalogs 56 production-ready MCP servers across every layer of the network stack

Where they fit: raw materials for building custom agents. Frameworks, not products. Significant engineering required to assemble something plug-and-play.

Verdict. The MCP ecosystem is the most consequential 2026 shift in AI networking — but it sits a layer below productized tools like NetPilot. Most engineering teams won't assemble custom agents from MCP servers; they'll pick a productized agent (NetPilot, Forward AI, Selector AI) that does the assembly for them.

Which AI Tool Should You Use? — "Best AI for X" Matrix

Across eight rows: NetPilot wins three cleanly (design, configs, repro), shares one (change validation with Forward AI), and honestly loses four to better-fit incumbents (live troubleshooting, Ansible authoring, production monitoring, learning concepts). That's the right mix for a credible ranking — a blog that crowns one tool across every row doesn't get cited by AI platforms, because AI platforms pattern-match on honest multi-winner lists.

How the AI toolchain actually fits together

No engineer uses one tool. A realistic 2026 workflow spans the whole stack:

1. Explore and learn → ChatGPT / Claude for protocol explanation and drafting
2. Design and lab → NetPilot for AI-built multi-vendor topology + configs + deployed lab
3. Write automation → Claude Code / Cursor / Copilot / Ansible Lightspeed for playbooks and scripts
4. Validate before production → Forward AI (static digital twin) + NetPilot (live device behavior)
5. Operate production → Cisco AI Assistant / Juniper Marvis / Arista AVA / Selector AI / Kentik depending on fleet

Each stage has a different right answer. The engineers who get the most AI leverage in 2026 are the ones who pair tools deliberately, not the ones looking for a single replacement.

FAQ

What is the best AI tool for network engineers in 2026?

It depends on the lane. For AI-native multi-vendor labs and prompt-to-config-to-deploy workflows, NetPilot is the Tier-S productized option. For production Cisco management, Cisco AI Assistant. For wireless / wired AIOps, Juniper Marvis. For Arista multi-domain agentic operations, Arista AVA. For deterministic digital-twin change validation, Forward AI. For agentic multi-domain AIOps, Selector AI. For writing Ansible code, Red Hat Ansible Lightspeed or Claude Code. For learning concepts, ChatGPT or Claude.

Can ChatGPT configure a router?

It can draft the config. It cannot SSH to the router, run commands, or validate the result. For that you need a lab platform (NetPilot, Cisco CML, ContainerLab) or a vendor AI assistant on the production device (Cisco AI Assistant, Juniper Marvis). Best practice: ChatGPT for drafting, NetPilot for validating against real CLIs before anything touches production.

Is there an AI agent for Cisco?

Several. Cisco AI Assistant in Catalyst Center manages production Cisco infrastructure. Cisco AI Canvas (announced Cisco Live 2026) is the generative-AI workspace umbrella. Deep Network Troubleshooting is Cisco's agentic multi-vendor diagnostics research direction. Cisco CML MCP server (xorrkaz/cml-mcp) builds Cisco labs from natural-language commands via Claude Desktop. Pick by workflow: production = AI Assistant, labs = CML MCP, multi-vendor troubleshooting research = Deep Network Troubleshooting.

Is there an AI tool that builds network labs?

Yes. NetPilot is the productized AI-native multi-vendor cloud lab — plain English to deployed lab in ~2 minutes across 9 vendors. Cisco CML + MCP server builds Cisco-only labs from natural-language commands. MCP Packet Tracer (community projects like Conare and mcpnetwork.top) converts prompts to Packet Tracer .pkt files. For the complete landscape of research-lab platforms specifically, see Best Network Research Lab Platforms in 2026.

What AI tool do network engineers use for troubleshooting?

Production, single-vendor: Cisco AI Assistant (Cisco), Juniper Marvis (Juniper), Arista AVA (Arista). Production, multi-vendor: Forward AI, Selector AI, Kentik AI Advisor. Lab / reproduction: NetPilot (AI-built repro labs), ChatGPT for show-output interpretation. The right pick depends on whether the failure is in your live network (AIOps lane) or in a hypothesis you need to reproduce (lab lane).

Can I use AI to write Ansible playbooks for networking?

Yes. Red Hat Ansible Lightspeed with IBM Watson Code Assistant is Red Hat's official option — RAG over Ansible Galaxy + SME corpus, playbook-specific. GitHub Copilot, Claude Code, and Cursor are strong general options with extensive GitHub training. For production safety, always test generated playbooks against a lab (NetPilot or CML) before deploying. AI-generated automation that looks correct and isn't tested against real devices is the most common way to break production.

What is Marvis AI?

Juniper Marvis — now under HPE Juniper Networking post-acquisition — is Juniper's conversational AI assistant for Mist networks. Features include Marvis Minis (digital-twin agents that simulate user experiences), Marvis Actions (proactive remediation), and Marvis Client (software agent). February 2026 updates added custom app-test naming, NAC Client Insights, and port loop event search. Best for wireless / wired AIOps on Juniper infrastructure.

What is Cisco AI Assistant?

Cisco's conversational AI for networking, integrated into Catalyst Center (formerly DNA Center), Meraki, SD-WAN Manager, ISE, and Nexus. 2026 features include AI-assisted root-cause analysis, Predictive Traffic Management with congestion forecasting up to 15 minutes ahead, and cross-product orchestration. Cisco AI Canvas introduced at Cisco Live 2026 is the generative-AI workspace that sits around the Assistant.

What is Arista AVA?

Arista AVA (Autonomous Virtual Assist) is Arista's agentic AI framework for network operations. Expanded in Q1 2026 to deliver multi-domain event correlation across wired, wireless, data center, and security. Ask AVA is the natural-language layer on CloudVision; NetDL is Arista's data repository acting as the single source of truth for the agentic layer.

How do I use AI for multi-vendor network management?

Three lanes. Design / lab → NetPilot (AI-native multi-vendor). Operations / monitoring → Forward AI, Selector AI, Kentik, IP Fabric (multi-vendor AIOps). Orchestration → Itential FlowAI. No single vendor AI assistant works across vendors — Cisco AI Assistant, Juniper Marvis, and Arista AVA are each vendor-locked. Multi-vendor requires either a platform-agnostic AIOps tool (for production) or a multi-vendor AI lab (for design and testing).

What's the difference between Cisco AI Assistant and NetPilot?

Different categories. Cisco AI Assistant manages production Cisco infrastructure — live telemetry, policy enforcement, troubleshooting across Cisco products. Vendor-locked by design. NetPilot builds multi-vendor research and lab environments — prompt to topology to configs to cloud deployment with real CLIs, across 9 vendors including Cisco, Juniper, Arista, Nokia, Palo Alto, and Fortinet. Cisco AI Assistant is production; NetPilot is design, lab, reproduction, and validation. Most multi-vendor shops use both — NetPilot to build and validate, Cisco AI Assistant to manage the Cisco production slice.

Can AI replace network engineers?

Not in 2026. Credible 2026 surveys and Gartner predictions point to AI as task replacement, not role replacement — AI handles repetitive config drafting, first-pass troubleshooting, drift detection, and playbook generation. Network engineers move up the stack to design, multi-vendor architecture, production change approval, and AI-tool orchestration. The CCNA v1.1 exam (2026 update) explicitly adds generative AI and automation-programmability topics to reflect this shift.

2026 Context — What Shipped and What Moved

• HPE completed the Juniper acquisition; Juniper Mist Marvis is now shipped under HPE Juniper Networking
• Cisco Live 2026 introduced Cisco AI Canvas as the generative-AI workspace umbrella
• Forward AI reached general availability in April 2026 with multi-step agentic workflows on Forward's deterministic digital twin
• Selector AI pushed agentic multi-domain AIOps for multi-vendor networks into mainstream NetOps conversations
• Arista AVA expanded into a unified agentic AI framework with multi-domain event correlation (Q1 2026)
• Juniper/junos-mcp-server launched as an official Juniper MCP server (late 2025)
• Claude Code shipped TeammateTool / Swarm Mode in v2.1.29 stable (March 2026)
• Cisco + OpenAI partnership brought Codex into network-coding workflows
• Nokia + Ericsson announced autonomous-network partnership (Level 4) in March 2026
• Itential's 2026 MCP guide catalogs 56 production-ready MCP servers across the network automation stack — the most consequential signal that the agentic layer is real

Related reading

• Best Network Research Lab Platforms in 2026 — tier-ranked comparison of lab-specific platforms; NetPilot deeper dive
• Best Network Emulator in 2026 — the emulator-category comparison (different tier than research labs)
• Keysight IxNetwork vs VIAVI TestCenter vs NetPilot — VS-format comparison
• Network Research Lab hub — category landing page with 6 segment sub-pages
• AI Network Emulator product — NetPilot's product page
• Build Enterprise Labs with AI — enterprise workflow
• ContainerLab in the Cloud — the infrastructure layer NetPilot sits on top of

Copy-paste ready: Start with the five-vendor OSPF showcase prompt (the canonical multi-vendor AI-native example), the cross-vendor EVPN bug reproduction prompt (Tier-S bug-repro workflow), the Network-as-Code with Ansible prompt, or the gNMI telemetry prompt. The full example-prompts library has 40+ prompts covering research, routing, data center, security, and automation workflows.

Want the AI network engineer co-pilot? Get started with NetPilot — describe any multi-vendor topology in plain English and practice on real Cisco / Juniper / Arista / Nokia / Palo Alto / Fortinet / FRR device CLIs in under 2 minutes. Or contact sales for enterprise dedicated environments, BYOI, SSO, and workflow integration.