Use Cases & Patterns

Practical ways to apply the AKIOS security cage in production.

Core patterns

Run untrusted/3rd-party agents safely with default-deny network/fs and audit.
Pre-flight in CI/CD: validate policies and costs before deploy.
Handle sensitive data: PII redaction + tamper-evident audit on every step.
Guarded tool execution: allowlist commands and paths, no ambient network.

Examples

1) Secure document processing (public sector)

name: "Secure Document Summary"
steps:
  - step: read_pdf
    agent: filesystem
    action: read
    parameters:
      path: "/inputs/application.txt"
  - step: analyze
    agent: llm
    action: complete
    parameters:
      model: "gpt-4"
      prompt: "Summarize this application: {{read_pdf.content}}"
  - step: save
    agent: filesystem
    action: write
    parameters:
      path: "/outputs/summary.txt"
      content: "{{analyze.text}}"

Why it fits: automatic PII redaction + signed audit trail.

2) API enrichment with budget guard (finance)

name: "KYC Enrichment"
steps:
  - step: fetch_data
    agent: http
    action: get
    parameters:
      url: "https://api.external-data.com/user/{{user_id}}"
  - step: risk_analysis
    agent: llm
    action: complete
    parameters:
      model: "claude-3.5-sonnet"
      prompt: "Analyze risk for this user data: {{fetch_data.body}}"
  - step: save_report
    agent: filesystem
    action: write
    parameters:
      path: "/reports/{{user_id}}_risk.json"
      content: "{{risk_analysis.text}}"

Why it fits: enforced budget/token limits + redaction on inputs/outputs.

3) Safe tool execution (DevOps)

name: "Log Analyzer"
steps:
  - step: run_script
    agent: tool_executor
    action: run
    parameters:
      command: "python3 /scripts/analyze_logs.py --input /logs/server.log"

Why it fits: command allowlist, default-deny network, bounded paths.