KKentik AI Advisor

Kentik says its data engine takes in a trillion telemetry points a day, and AI Advisor is the agent that now sits on top of that pile. Ask it why latency to a cloud region jumped or where transit spend can be cut, and it builds a plan, runs queries across flow, device, cloud and synthetic data, and walks through its reasoning. It suits service providers and large network teams that already feed Kentik. It will not touch router configurations, and it is not sold on its own.

The entry ticket is the Kentik platform itself: Pro starts at $2,000 a month billed annually, with a 30-day free trial and Premier on quote. We rate it 8.4/10 on our scale. We have not used Kentik ourselves; this review is based on Kentik’s product documentation, pricing page and launch announcement, plus Gartner Peer Insights ratings.

A Published Entry Price and Clear Limits on the Agent

Kentik is one of the few network observability vendors that publishes an entry price, and its documentation is unusually specific about what the agent can and cannot touch.

Two facts matter most for a buyer. First, the agent is only as smart as the data underneath it, and Kentik’s pitch rests on the scale and breadth of that data: flow records, device metrics, cloud flow logs, BGP state and synthetic tests in one store. A team that already pays to send that telemetry to Kentik gets the agent on top of it; a team that does not is really buying a network observability platform first.

Second, the read-only design is a deliberate limit. Kentik’s docs say every AI query is read-only and the agent cannot change device configurations. For carriers wary of an AI touching production routers, that is a selling point. For teams looking for closed-loop remediation, it means AI Advisor stops at diagnosis and recommendation.

Plain-Language Prompts Turned Into Planned Network Queries

AI Advisor is a conversational agent inside the Kentik portal. According to Kentik’s knowledge base, it interprets a natural-language prompt, plans the steps needed to answer it, runs tools and shows its progress in the conversation in real time.

Where Engineers Start It

• Full page: the AI Advisor link in the main navigation opens a dedicated workspace.
• Overlay: an Ask button in most page headers, including device details, opens the agent beside the current view.
• From an alert: Investigate with AI Advisor on alert pages hands the alert straight to the agent as a starting point.

What It Can Reach

Kentik lists about 50 tools, and says the set is not limited to those. They fall into a few groups:

• Telemetry: device metrics such as CPU, memory, interface counters and BGP state; flow data from NetFlow, IPFIX, sFlow and VPC flow logs; syslog and SNMP traps.
• Inventory and topology: device, interface and site inventories, device connections, configuration history and diffs, and cloud inventories for AWS, Azure, Google Cloud and Oracle Cloud.
• Alerts and mitigation: alerts, alert statistics, mitigation events and BGP mitigation announcements, which supports DDoS investigation.
• Active diagnostics: read-only show commands on devices, ping and traceroute from external agents, DNS lookups and Cloud Pathfinder.
• Reference data: Kentik Market Intelligence, PeeringDB, ASN and Whois lookups, network provider lookups and Kentik’s own knowledge base.

Through agent-to-agent (A2A) access it can also call other Kentik agents, including a Connectivity Cost Agent and a Kentik Market Intelligence agent. Kentik also offers an AI Advisor MCP server for use from other AI tools.

Teaching It About Your Network

Two settings, both managed by a super administrator, shape answers. Runbooks are Markdown troubleshooting recipes that can be assigned to alert policies, so the agent follows the same diagnostic path each time a given alert fires. Custom Network Context is up to 100K characters describing architecture, IP schemes, naming conventions, maintenance windows and critical applications, and it applies to every conversation in the organization.

Typical Jobs

Kentik’s launch announcement groups the work into four areas: cost optimization across VPC and transit spend, peering and interconnects; capacity planning from utilization trends and run-out forecasts; troubleshooting and DDoS investigation by correlating flow, device and cloud data; and applying internal runbooks so answers reflect how a team actually operates.

Read-Only Access, So Device Configs Stay Untouched

Kentik’s docs state plainly that AI Advisor is read-only and cannot change network device configurations. It can write to Kentik itself in limited ways, such as creating or updating sites and runbooks. Every router change, mitigation decision or capacity order stays with an engineer. Setup is light: a super administrator enables Kentik AI for the organization, users refresh the portal, and the agent is available. AI features are off by default.

No Separate Fee Beyond the Kentik Platform

AI Advisor is not priced on its own. It runs inside the Kentik platform, and Kentik publishes its entry plan on its pricing page.

A Regional ISP’s 12-Person Network Team on Pro

The scenarios below use Kentik’s published Pro price; team sizes and volumes are our own illustrative assumptions.

• A regional ISP with a 12-person network team: Pro at $24,000 a year covers all 12 users within the 50-user cap, or $2,000 per engineer per year. A three-year commitment at list price comes to $72,000 before add-ons such as Service Provider Analytics.
• A carrier with 80 operations staff: that passes the Pro limit of 50 users, so the team moves to Premier on a custom quote.
• Device monitoring: Pro includes 25 NMS devices. A provider that wants device metrics for hundreds of routers so the agent can query them should price the extra NMS capacity before signing.

Hidden or variable costs to plan for: add-ons for DDoS protection and service provider analytics, flow volume above the included 1,000 flows per second, professional services such as RapidStart onboarding, and staff time to write runbooks and network context. International pricing is quoted separately.

Kentik AI Advisor Compared With Selector AI, Nokia Autonomous Network Fabric and B-Yond AGILITY

Network agents for telecom split by what they sit on: an observability data platform, an equipment vendor’s management stack, or packet-level diagnostics.

Kentik is the only one of the four with a published price, which makes it the easiest to budget. Selector goes further toward action, with agents that stage fixes for approval. Nokia makes sense for operators already deep in its equipment and software, and B-Yond is a specialist for mobile core and RAN protocol failures. Browse more in our AI agents for telecom and IT and DevOps guides.

Service Providers Already Sending Telemetry to Kentik

• Service providers already on Kentik: the agent works on flow, BGP and device data you already send, so the added value comes without a new data project.
• Network teams short on senior engineers: Kentik pitches the agent at teams facing talent shortages, and runbooks let less experienced staff follow a senior engineer’s diagnostic path.
• Teams chasing transit and cloud costs: cost analysis across VPC, transit, peering and interconnects is one of the four jobs Kentik designed it for.
• Operators investigating DDoS events: the agent can read alerts, mitigation events and BGP mitigation announcements together.

Teams Wanting Auto-Remediation and Small Networks

• Teams that want automatic remediation: it cannot change device configurations, so look at agents that stage or apply fixes.
• Small networks: a $24,000 a year annual contract is a lot for a handful of routers.
• Operators whose problems live in the mobile core or RAN protocol layer: packet-level tools such as B-Yond focus on that work.
• Buyers who need billing or customer care agents: AI Advisor is a network operations tool; BSS-focused platforms such as Amdocs aOS cover care and billing.

Checking Answers and Keeping Context Current

• Answer checks: Kentik’s own docs say the agent can make mistakes, state incorrect information confidently and vary its answers to the same prompt over time. Check its evidence before acting.
• Conversation memory: context lasts for one conversation. Only runbooks and Custom Network Context carry over, so those need upkeep.
• Data dependence: answers rely on telemetry in Kentik. Gaps in flow export or device polling become gaps in diagnosis.
• Third-party models: prompts go to OpenAI, Anthropic, Google Cloud or AWS-hosted models. Kentik says partners may not train on or keep the data, apart from short abuse monitoring, but regulated operators should review this.
• Annual terms: Pro and Premier are annual contracts with no monthly option.
• Plan limits: Pro caps users at 50 and API queries at 50 an hour, which larger teams can outgrow.

A 2025 Launch Aimed at Short-Staffed Infrastructure Teams

• On November 18, 2025, Kentik launched AI Advisor, pitching it at infrastructure teams facing talent shortages and contrasting it with traditional AIOps that only correlates alerts it receives.

A Fit for Large Network Teams Already on Kentik

For service providers and large network teams already sending telemetry to Kentik, yes. Kentik AI Advisor scores 8.4/10 on our scale: strong on data reach and user sentiment, with room to grow on remediation and entry price. It runs real multi-step investigations across flow, device, cloud and synthetic data, and its read-only design and no-training policy make it an easy agent to approve in a cautious carrier.

For teams not on Kentik, the decision is really about the platform. At $24,000 a year for Pro, you are buying network observability with an agent included, not an agent alone. Use the 30-day trial to point the agent at real alerts and check whether its explanations match what your engineers would have found.

Router Config Changes, Models and Agent Cost

How Much Does Kentik AI Advisor Cost?

AI Advisor comes with the Kentik platform rather than a separate price. Kentik Pro starts at $2,000 a month billed annually, or $24,000 a year in the US, and Premier is quoted. Add-ons such as Service Provider Analytics are extra.

Can Kentik AI Advisor Change Router Configurations?

No. Kentik’s documentation says it is read-only on network devices. It can run show commands, ping and traceroute, and it can create or update sites and runbooks inside Kentik, but configuration changes stay with your engineers.

Which AI Models Does Kentik Use?

Kentik lists OpenAI models hosted by OpenAI, Anthropic models hosted by Anthropic or through Amazon Bedrock, and models from Anthropic, Google and others on Google Cloud Vertex AI. It says customer data is not used to train models.

How Is Kentik AI Advisor Different From Selector AI?

Both correlate network data to find root causes. Kentik’s agent is read-only and comes with a published platform price. Selector’s agents can stage fixes, such as a configuration rollback, for an engineer to approve, and Selector sells on custom quotes.

What Is Outside Kentik AI Advisor’s Scope?

It does not change device configurations, carry memory beyond runbooks and network context, or work on data outside Kentik. Kentik also notes it can state incorrect information confidently, so engineers should check its evidence.