Response Rewriter

The response-rewriter policy modifies AI provider responses before they are returned to the caller, enabling disclaimer injection, content redaction, and regex-based text transformations while optionally preserving JSON structure.

Use this page when

• You need to append disclaimers, redact brand names, or transform AI responses before they reach callers.
• You are injecting compliance disclaimers conditionally (e.g., financial disclaimer only when "investment" appears).
• You want regex-based response transformations while preserving JSON structure.

Primary audience

• Primary: AI Agents, Technical Engineers
• Secondary: Technical Leaders

Configuration

policy:
  response-rewriter:
    preserve_structure: true
    rules:
    - name: add-disclaimer
      pattern: "$"
      replacement: |2-


        ---
        _This response was generated by AI and may contain errors._
      position: append
    - name: redact-competitor-names
      pattern: '\b(CompetitorA|CompetitorB|CompetitorC)\b'
      replacement: "[COMPETITOR]"
      position: replace
pack:
  name: response-rewriter-example-1
  version: 1.0.0
  enabled: true
policies:
  chain:
  - response-rewriter

Fields

Field	Type	Description	Default
preserve_structure	bool	Preserve JSON structure in responses. When enabled, rewrite rules are applied only to string-valued text fields, leaving JSON keys, numbers, and booleans untouched.	true
rules	RewriteRule[]	Ordered list of regex-based rewrite rules applied to response content. Rules execute in declaration order.	[]
rules[].name	string	Required. Unique identifier for the rule, used in audit logs.	--
rules[].pattern	string	Regular expression pattern to match in the response content. Supports capture groups.	--
rules[].replacement	string	Replacement string. Use $1, $2, etc. to reference capture groups from the pattern.	--
rules[].position	string	How the replacement is applied: "replace" substitutes the match, "prepend" inserts before the match, "append" inserts after the match.	"replace"
rules[].condition	string	Optional substring guard. The rule only applies if this substring is present in the response content.	--

Use Cases

Adding Compliance Disclaimers to Responses

Append a compliance disclaimer to every AI response in a regulated environment.

pack:
  name: "compliance-disclaimers"
  version: "0.1.0"
  enabled: true

policies:
  chain:
    - safety-filter
    - response-rewriter
    - audit-logger

policy:
  safety-filter:
    action: "block"

  response-rewriter:
    preserve_structure: true
    rules:
      - name: "general-disclaimer"
        pattern: "$"
        replacement: '\n\n_Disclaimer: This AI-generated response is for informational purposes only and does not constitute professional advice._'
        position: "append"
      - name: "financial-disclaimer"
        pattern: "$"
        replacement: '\n_This is not financial advice. Consult a qualified financial advisor._'
        position: "append"
        condition: "investment"

  audit-logger:
    retention_days: 365

Redacting Brand Names

Replace competitor or internal brand names in responses with generic placeholders.

pack:
  name: "brand-redaction"
  version: "0.1.0"
  enabled: true

policies:
  chain:
    - response-rewriter

policy:
  response-rewriter:
    preserve_structure: true
    rules:
      - name: "redact-competitors"
        pattern: '\b(Acme Corp|Globex|Initech|Umbrella Corp)\b'
        replacement: "[THIRD_PARTY]"
        position: "replace"
      - name: "redact-internal-codenames"
        pattern: '\b(Project (Phoenix|Titan|Orion))\b'
        replacement: "[INTERNAL_PROJECT]"
        position: "replace"

Injecting Safety Warnings

Prepend a safety warning to responses that mention dangerous activities.

pack:
  name: "safety-warnings"
  version: "0.1.0"
  enabled: true

policies:
  chain:
    - response-rewriter
    - audit-logger

policy:
  response-rewriter:
    preserve_structure: false
    rules:
      - name: "chemical-warning"
        pattern: "^"
        replacement: '[WARN] SAFETY WARNING: The following information involves hazardous materials. Follow all applicable safety regulations.\n\n'
        position: "prepend"
        condition: "chemical"
      - name: "electrical-warning"
        pattern: "^"
        replacement: '[WARN] SAFETY WARNING: Electrical work should only be performed by qualified professionals.\n\n'
        position: "prepend"
        condition: "voltage"

  audit-logger:
    retention_days: 90

How It Works

1. Response interception -- The gateway intercepts the response from the upstream AI provider before returning it to the caller.
2. Structure detection -- If preserve_structure is enabled, the gateway identifies the response format (JSON, streaming SSE, plain text) and extracts text fields for rule application.
3. Rule evaluation -- Rules execute sequentially in declaration order. Each rule's condition (if set) is checked first; if the substring is not present in the response content, the rule is skipped.
4. Pattern matching and replacement -- The pattern regex is applied to the extracted text content. Matches are processed according to position: "replace" substitutes, "prepend" inserts before, "append" inserts after.
5. Structure reassembly -- When preserve_structure is enabled, modified text fields are placed back into the original response structure, preserving JSON formatting, streaming chunk boundaries, and metadata.

Best Practices

• Keep preserve_structure enabled for API consumers. Disabling it on JSON responses can break downstream parsers that depend on the response schema.
• Place response-rewriter after content-filtering policies. Apply safety filters and content checks first, then rewrite the approved response. This avoids rewriting content that would have been blocked.
• Use condition guards for context-specific disclaimers. Adding financial disclaimers only when the response mentions investment topics reduces noise and improves user experience.
• Avoid overlapping patterns. If multiple rules can match the same text, ensure they are ordered so earlier rules do not create text that triggers later rules unintentionally.
• Test with streaming responses. Regex patterns that span chunk boundaries may not match in streaming mode. Prefer patterns that match within a single text segment.

For AI systems

• Canonical terms: Keeptrusts, response-rewriter, preserve_structure, rules, pattern, replacement, position, condition, append, replace, prepend
• Config/command names: policy.response-rewriter, preserve_structure, rules[], rules[].pattern (regex), rules[].replacement, rules[].position (replace/prepend/append), rules[].condition
• Best next pages: Request Rewriter, Financial Compliance, Safety Filter

For engineers

• Prerequisites: Regex patterns for content to transform. Disclaimer text approved by compliance. Understanding of JSON structure preservation requirements.
• Validation: Send a request that generates a response containing your target patterns and verify transformations. Test preserve_structure: true with JSON responses to confirm keys/numbers are untouched.
• Key commands: kt policy lint, kt gateway run, kt events tail

For leaders

• Governance: Response rewriting ensures every AI output includes required disclaimers and excludes sensitive brand/competitor references — enforced at the infrastructure layer.
• Cost: Local regex processing with no external calls. Negligible per-request overhead.
• Rollout: Start with universal disclaimers (append to all responses). Add conditional rules (disclaimer only when specific keywords appear) as you refine requirements.

Next steps

• Request Rewriter — Transform outbound requests
• Financial Compliance — Finance-specific disclaimers
• Healthcare Compliance — Medical disclaimers
• Quality Scorer — Quality gates on transformed responses