Reza Koohrangpour describes Selector's answer to the hybrid observability problem as three capabilities built on a horizontal architecture: collection, correlation, and collaboration. Selector collects data from more than 300 heterogeneous sources into one unified store, covering cloud accounts, regions, syslog, VPCs, gateways, and infrastructure and network signals, so there is no separate cloud view and network view. Because latency spikes, application degradation, and routing failures are rarely independent, Selector uses machine learning, causation, and correlation to group related signals into a single incident with mapped dependencies. Collaboration then delivers that incident, with context and evidence attached, in plain language to the right people in tools such as Slack and Teams. The platform runs as Kubernetes microservices and scales horizontally. Its collection layer ingests raw telemetry such as SNMP, NetFlow, and gNMI; data from third-party tools like Splunk and ThousandEyes; cloud metrics and logs; and contextual sources such as topology, CMDB records, and knowledge bases, with Selector-provided collectors deployable wherever the data lives. A programmable "data hypervisor" layer, configured with YAML and DevOps workflows, normalizes and deduplicates data, suppresses noise, and enriches it with business logic, maintenance windows, and provider information.

Delegate questions drew out several further details, largely answered by Koohrangpour's colleague John Hientz. Selector supports both reactive troubleshooting and proactive use, including forecasting when bandwidth will reach saturation or when degrading optics will fail. Some customers use it as a network-focused digital twin to simulate the effect of changes before running a method of procedure. On data volume, Selector positions itself as an operational tool rather than a long-term log archive, retaining raw data for roughly 30 to 60 days while keeping normalized data and detected anomalies much longer across multiple purpose-built data stores. It can be deployed on-premises or in a customer's cloud for regulated industries such as banking and utilities, or consumed as SaaS. Asked about consistency when using frontier models, the team explained that Selector front-loads the work of normalization, enrichment, and anomaly detection, backed by more than 13 patents, so models do not reason over raw data. Instead, the models translate user questions into queries against Selector's curated data lake and present the results, which keeps answers grounded. Customers can bring their own model, with Gemini, ChatGPT, and Claude all cited as options.

Koohrangpour then described the upper layers of the stack. Knowledge services feed enriched data into the correlation engine and topology graph, and collaboration services expose insights through an MCP layer that governs which Selector tools agents can use, such as cloud infrastructure, VPN, BGP, and path tracing, while allowing secure connections to external MCP servers like ServiceNow, Slack, and Teams. On top of this sits Selector Foundry, the agentic runtime launched the day before the presentation. It includes Rosetta, a conversational interface that orchestrates specialized agents such as Loom for investigation and Artist for building dashboards from plain-English requests, along with agents that can create new agents to automate repeatable workflows triggered by specific events. Koohrangpour closed by stressing that Selector treats cloud as simply another data source, so AWS, Azure, and Google Cloud events, logs, and flows sit alongside SNMP and NetFlow data and share the same topology, correlation, and agents. That unified approach lets operators follow a service path end to end, from on-premises VLANs through interconnects and hubs into cloud VPCs and transit gateways, and reach root cause without switching tools.

Presented by Reza Koohrangpour, Head of Product & Solutions Marketing, Agentic AI, Selector, and John Heintz, Senior Global Systems Engineering Director, Selector. Recorded live at Cloud Field Day in Santa Clara, California, on September 23rd, 2026. Watch the entire presentation https://techfieldday.com/appearance/selector-ai-presents-at-cloud-field-day-26/ or visit https://techfieldday.com/event/cfd26/ or https://www.selector.ai/ for more information.