Skip to content

Analyzer (default)

The analyzer is the default netops sub-agent: a read-only change-plan drafter. It gathers device facts from the snapshot database via SQL, then writes a vendor-specific change plan — CLI per device, rollback, post-checks, and a risk classification — to exports/change_plans/. It is read-only on live devices: it reads the DB, it never connects to or configures a device.

The analyzer also owns two other jobs that were folded into it:

  • Investigation reports (Workflow D) — an audit / "why" write-up, exported to exports/reports/.
  • Snapshot diff & drift (absorbed from the retired ops-diff sub-agent) — compare configs, snapshots, SQL state, and topology between two points in time.
  • L2 topology what-if — blast-radius over the graph topology (inspect_blast_radius), not a Batfish config simulation.

What-if / compatibility is a different sub-agent

Config-layer what-if — "would BGP come up if…", reachability, route-map policy — is the Batfish-backed simulator. The analyzer delegates there via task("simulator", "…"). Topology queries and recipe discovery are the topology sub-agent.

When to use

  • "plan / add / change / modify / 变更 a new eBGP session between R1 and R4"
  • "write an investigation report on why R3 lost its routes"
  • "what changed since yesterday's snapshot?" (drift / diff)
  • "what's the blast radius if the R1–R2 link goes down?"

Tools

Tool When to use
describe_table Inspect a view/table's columns before querying
inspect_devices Structured device facts — pass devices=[] to discover all
inspect_interfaces Structured interface facts for the target nodes
inspect_blast_radius L2 graph-topology what-if (blast radius) — not Batfish
generate_change_plan Fat tool — full multi-device BFS change/upgrade plan in ONE call
query_evidence Unified text search across syslog / command output / config
diff_configs Compare device configs between two snapshots
diff_snapshots Compare two snapshots (state / topology drift)
format_and_export Render + export the plan or report to exports/

The analyzer never holds execute_cli or any write tool — it produces a plan, it does not apply it.

Example

olav --agent netops "How do I add an eBGP session between R1 and R4?"

The orchestrator routes to the analyzer, which reads the relevant device facts (inspect_devices, query_evidence), calls generate_change_plan, and writes a Markdown plan — current state, proposed config per device, rollback, post-checks, and risk — to exports/change_plans/. A config-layer what-if is delegated onward:

task("simulator", "would the eBGP session between R1 and R4 come up with this config?")