Selector AI

No page on the site addresses this.

Create a dedicated 'Getting Started' or 'POC Requirements' page that explicitly lists minimum data source requirements for a 30-day POC, including whether SNMP, syslog, and ticketing system integration are required vs. optional for demonstrating meaningful correlation.

Where we looked: /resource/networking-field-day-30-deep-dive-into-selector-ai, /blog/breaking-down-silos-with-correlation-and-context, /resource/networking-field-day-30-deep-dive-into-selector-ai

No page on the site addresses this.

The chunks mention that 'customization is available' and customers can 'request additional data to be included' during managed service engagement, but do not specify whether custom field mappings are preserved bidirectionally or if there's a limit to how many custom fields can be mapped. Need explicit documentation on ServiceNow field mapping capabilities, custom field support limits, and whether bidirectional sync preserves custom form configurations.

Where we looked: /resource/networking-field-day-35-selector-ai-alerting-discussion-with-nitin-kumar, /blog/selector-mcp-and-the-future-of-modular-automation

No page on the site addresses this.

Create a dedicated data retention policy documentation page that specifies default hot/warm/cold storage tiers, searchable retention periods, and whether customers can configure custom retention splits. The current content only mentions that 'retention policies and how long something is stored' are controlled by admins without exposing defaults or configurability to users.

Where we looked: /resource/networking-field-day-30-deep-dive-into-selector-ai, /docs/getting-started/overview-selector-v1

No page on the site addresses this.

The 'intelligent capture' feature is mentioned in marketing copy but lacks technical specifications. Add a technical documentation page for Packet Copilot covering: disk footprint calculations for common link speeds (1Gbps, 10Gbps, 100Gbps), default capture mode (full packets vs headers), configurable retention policies, and compression ratios.

Where we looked: /solutions/synthetic-monitoring-and-analytics, /solutions/wi-fi-device-analytics, /network-llm-copilot

Selector uses ML-driven techniques for alert deduplication and root cause analysis, but the specific mechanism for determining 'same root cause' and whether there are configurable similarity thresholds is not fully documented in the retrieved chunks. The system uses 'collaborative filtering' clustering to group related events into clusters, and forms 'correlation trees' where connected edges indicate related events. The correlation is described as both 'temporal and contextual.' However, no explicit mention of user-configurable similarity thresholds was found. One chunk mentions 'Composite alerting enables customers to leverage a combination of alerting conditions to further filter and tune the types of alerts,' suggesting some configurability exists, but this does not directly address deduplication thresholds.

Add technical documentation explaining the deduplication algorithm specifics, including: (1) the exact criteria used to determine 'same root cause' (e.g., topology proximity, temporal window, event type similarity), (2) whether similarity thresholds are configurable by users or administrators, and (3) if ML models can be tuned or overridden with custom rules.

Where we looked: /docs/getting-started/overview-selector-v1, /resource/networking-field-day-30-deep-dive-into-selector-ai, /resource/networking-field-day-35-selector-ai-alerting-discussion-with-nitin-kumar, /blog/breaking-down-silos-with-correlation-and-context, /docs/getting-started/overview-selector-v1

Selector offers a true on-premises deployment option where 'the customer hosts the VMs and Disks that the Selector platform uses' and 'Selector uses LongHorn to replicate the data stores across the cluster.' The platform is built on Kubernetes, enabling deployment 'on-premises, off-premises, or in the cloud.' However, the chunks do not explicitly clarify whether the 'hybrid' deployment model involves cloud-based processing or if it's purely a data collector with local processing. The company describes deployment options as 'SaaS, hybrid, or on-prem' but does not define the technical architecture of the hybrid model specifically.

Create a dedicated deployment architecture page that explicitly defines what 'hybrid' means technically—specifically whether processing occurs in Selector's SaaS or remains on-premises, and what data (if any) leaves the customer environment in each deployment model.

Where we looked: /docs/getting-started/overview-selector-v1, /blog/why-selectors-sres-chose-selector-for-kubernetes-and-multi-cloud-application-observability, /executive-leadership, /resource/networking-field-day-30-deep-dive-into-selector-ai, /resource/networking-field-day-35-selector-ai-alerting-discussion-with-nitin-kumar

Selector supports maintenance windows with alert suppression for device groups, and has enhanced this feature with a 'strict_match' tag for more granular control. However, the chunks do not explicitly confirm whether these maintenance windows can be scheduled as recurring (e.g., every Tuesday 2-4am) or if they require manual creation each time. The release notes describe how maintenance windows work and how alert suppression behaves when devices are under maintenance, but do not specify the scheduling mechanism (recurring vs. one-time).

Add explicit documentation about whether maintenance windows support recurring schedules (cron-like expressions) or require manual creation for each occurrence. The current docs describe the suppression logic but not the scheduling interface.

Where we looked: /docs/relnotes/selector-relnotes-apr2026, /docs/relnotes/selector-relnotes-apr2026, /docs/relnotes/selector-relnotes-apr2026

Selector AI uses a hybrid training approach. The bulk of model training and fine-tuning happens in Selector Cloud, with models published to customers "every week or every other week" [5]. For customer-specific adaptation, there is "a thin layer of fine-tuning, which takes just a couple of hours" [5]. In another context, local training for LLM fine-tuning takes "about half an hour to an hour" once data is available [13]. However, the chunks do not specify: (1) whether customers can force a manual retrain of the anomaly detection models after events like a network redesign, or (2) whether they must wait for the scheduled weekly/bi-weekly cycle. The feedback loop described for baselining allows human feedback on alerts, but this appears to be for pruning false positives rather than forcing model retraining [8].

Add explicit documentation on whether customers can trigger manual retraining of anomaly detection models, or if retraining is only on a fixed schedule. Clarify the process for handling major environment changes like network redesigns.

Where we looked: /resource/networking-field-day-35-selector-ai-and-the-workings-of-an-llm, /resource/networking-field-day-35-selector-ai-and-the-workings-of-an-llm, /resource/networking-field-day-30-deep-dive-into-selector-ai

Yes, Selector is SOC 2 Type II certified. According to their newsroom announcement, 'Selector also received its SOC 2 Type II Attestation, demonstrating the company's ability to implement critical security policies and prove compliance. The SOC 2 Type II audit was also performed by Prescient Security, who verified that all necessary safeguards were in place to protect customer data and that the safeguards were operational.' Their documentation also confirms 'SOC 2 Type 2 & ISO 27001 Certification: We engage an independent, third-party auditor (approved by the AICPA) to conduct comprehensive annual assessments of our SaaS infrastructure, development processes, and overall Information Security Management System (ISMS).' However, the chunks do not specify whether the SOC 2 Type II report can be obtained under NDA before procurement.

Add explicit information about how prospects can request the SOC 2 Type II report under NDA, including contact information or a security questionnaire/request process on the security documentation page.

Where we looked: /newsroom/selector-onboards-ciso-to-strengthen-security-posture-and-receives-key-security-certifications, /docs/security/system-security-v1

Selector AI's support SLA specifies: Severity 1 (service down/major malfunction): 1 hour response time via Priority Contact (phone); Severity 2 (critical loss of functionality): 2 hours response time via Priority Contact (phone); Severity 3 (moderate loss): 4 Business Hours via Standard Contact (Support Portal); Severity 4 (minor issues/how-to): 3 Business Days via Standard Contact. However, the chunks do not explicitly state whether these SLAs include weekends or if 'Business Hours' excludes weekends.

The SLA table should explicitly clarify whether response times are calendar hours or business hours for Severity 1-2, and whether weekend coverage is included. The current 'Business Hours' notation for Severity 3-4 suggests weekdays only, but Severity 1-2's '1 Hour'/'2 Hours' without qualification leaves weekend coverage ambiguous.

Where we looked: /docs/support/selector-support-v1