Network Copilot details background agents in 1.7.0

Network Copilot 1.7.0 adds Background Agents, a feature for scheduling AI jobs on CRON intervals or as one-time tasks without user action. The update centers on unattended execution for routine network operations, with access controls, isolated runs, notifications, and run-level logging for enterprise teams.

Product Update

Background Agents use Celery and RedBeat on top of Network Copilot’s existing message queues and data layers. The same agents, tools, and connectors used in interactive chat can now run in the background for diagnostics, reporting, health checks, and ticket triage.

The release says scheduled jobs can run from chat or a form, with the backend validating permissions and saving jobs before execution. RedBeat and Celery Beat handle timing, Celery workers run the jobs, and overlapping executions are skipped and logged.

Operational Impact

The blog says the feature addresses four gaps in interactive-only execution: manual diagnostics, missed off-hours anomalies, recurring reporting that cannot be delegated, and limited compute governance. It also says background execution is project-scoped and role-gated, so jobs do not depend on the original creator remaining available.

Delivery is separated into in-app notifications and optional Slack messages, which the post says do not depend on each other. Each run also creates its own result tab, keeping background output separate from live chat sessions.

Key Findings

The post lists scheduled CRON runs, one-time background execution, skip-if-running behavior, isolated result tabs, independent Slack and bell notifications, and per-run receipts. The receipts record token use, duration, iteration counts, connector usage, and system metadata.

Example use cases include BigQuery CPU anomaly sweeps, weekly device-inventory reports, BCM node health checks, and overnight Zendesk incident summaries. The blog says these tasks can be completed before a shift begins.

Technical Breakdown

Under load, the blog says a 30-minute test on a four-worker deployment handled 190 concurrent scheduled jobs. It reports 434 runs observed, a 99.5% success rate, p50 latency of 9.1 seconds, p95 latency of 14.9 seconds, flat memory near 1.5 GiB, and balanced project fairness.

The post says the system had room for about 400 concurrent jobs before p95 latency would double. Planned additions include webhook and data-connector triggers, dynamic prompt templating, context memoization, and per-role token quotas with auto-pause.

Network Copilot Background Agents are positioned as a way to run repeated network tasks without manual prompting, while keeping execution bounded, logged, and separated from interactive work. This Blog Signals brief is a fact-based summary of the vendor blog.

Blog post, originally published by Rishab Haltore at aviznetworks.com.