Skip to content

Learner

The learner does parser-layer learning for CLI output. When the stock ntc-templates can't parse a device's output, the learner produces a persistent parser — a TextFSM template or a Python def parse(raw) — so every subsequent pipeline run parses that command automatically.

Invoke it via /learn_cmd: it learns once, freezes the parser, and every later collection picks it up without restart.

When to use

  • "The parser missed show ip bgp community — teach it"
  • "No template exists for this vendor's show version output"
  • "Batch-learn parsers for these commands that came back unparsed"

Tools

Tool When to use
learn_commands Batch mode — learn parsers for several unparsed commands at once
cmd_learn Interactive single-command learning (/learn_cmd)

Example

olav --agent netops "/learn_cmd 'show ip bgp community' --device R1"

The orchestrator routes to the learner, which analyzes the raw output, generates a TextFSM (or Python) parser, validates it, and freezes it. From then on the collection pipeline parses that command with no further action.

task("learner", "learn parsers for the commands that came back raw in the last snapshot")