Keyword Signal

Overview

keyword matches explicit words and phrases in the request. Define keyword rules under routing.signals.keywords.

This family is heuristic: it routes from configured words, phrases, or lightweight retrieval methods instead of learned intent classifiers.

Key Advantages

• Gives deterministic routing for obvious lexical cases.
• Supports simple regex-style matching and stronger BM25 or n-gram variants.
• Is easy to audit because the trigger phrases are explicit.
• Often provides the fastest path to a useful routing graph.

What Problem Does It Solve?

Some routes do not need a full classifier. They just need to recognize stable words such as billing, password reset, or urgent support.

keyword solves that by turning lexical matches into reusable named signals instead of scattering string checks across decisions.

When to Use

Use keyword when:

• lexical cues are stable and high-signal
• you want deterministic routing for support or policy keywords
• you need a low-latency first pass before learned signals
• prompt wording matters more than semantic paraphrase coverage

Configuration

routing:
  signals:
    keywords:
      - name: code_keywords
        operator: OR
        method: bm25
        keywords: ["code", "function", "debug", "algorithm", "refactor"]
        bm25_threshold: 0.1
        case_sensitive: false
      - name: urgent_keywords
        operator: OR
        method: ngram
        keywords: ["urgent", "immediate", "asap", "emergency"]
        ngram_threshold: 0.4
        ngram_arity: 3
        case_sensitive: false

Use plain keyword lists for simple matching, then add method: bm25 or method: ngram when exact text matching becomes too brittle.

Dependencies and Limitations

Keyword rules require no model, but they are sensitive to wording and can be triggered intentionally. Use them for routing hints, not security boundaries, and test false positives across supported languages. See complete examples for nlp.yaml and regex.yaml.