Configuration Reference

Gideon is configured using a central gideon.config.yaml file. This file controls everything from threat intelligence sources to safety guardrails and output formats.

Sources Configuration

Each security intelligence source can be enabled or disabled and tuned for performance.

sources:
  nvd:
    enabled: true
    base_url: "https://services.nvd.nist.gov/rest/json/cves/2.0"
    rate_limit: 5     # requests per second
    cache_ttl: 900    # 15 minutes
  virustotal:
    enabled: true
    rate_limit: 4     # requests per minute (free tier)

Agent Governance

Configure how Gideon monitors and controls its own behavior.

governance:
  enabled: true
  monitoring:
    behavioral_profiling: true
    anomaly_sensitivity: 0.7
  access_control:
    require_justification: true

Safety & Redaction

Gideon is designed with a Defensive-Only posture. These settings ensure it never generates or processes offensive content.

safety:
  defensive_mode: true      # Strictly blocks offensive capabilities
  block_offensive: true     # Prevents generation of exploit code
  require_explicit_auth: false

Sensitive Data Redaction

Gideon automatically scrubs sensitive data (API keys, passwords, tokens) from its outputs using regex patterns.

redaction:
  enabled: true
  patterns:
    - "(?i)(api[_-]?key|token|password|secret)\\s*[:=]\\s*['\"]?([^'\"\\s]{8,})"

AI Guardrails (NVIDIA NeMo)

Gideon uses NVIDIA NeMo Guardrails to enforce topical boundaries.

guardrails:
  enabled: true
  input_rails:
    jailbreak_detection: true # Blocks prompt injection/jailbreaking
    topic_control: true       # Ensures conversation stays on security
  allowed_topics:
    - cybersecurity
    - vulnerability analysis
    - incident response
  blocked_topics:
    - offensive security
    - malware development

OpenClaw Sentinel

Configure the sidecar security platform for OpenClaw AI agents. All settings live under the openclaw: key.

openclaw:
  enabled: true
  gateway:
    gateway_url: "ws://127.0.0.1:18789"  # OpenClaw WebSocket endpoint
    bind_mode: localhost                   # Expected bind mode
    openclaw_home: "~/.openclaw"           # OpenClaw data directory
  sentinel:
    enabled: true
    behavioral_profiling: true             # Build per-session behavior baselines
    detect_cve_2026_25253: true            # CVE-2026-25253 kill chain tracking
    anomaly_sensitivity: 0.7               # Behavioral anomaly threshold
    kill_chain_threshold: 2                # Stages before critical alert
  skill_scanner:
    enabled: true
    block_critical: true                   # Auto-block critical-risk skills
    check_typosquatting: true              # Detect name impersonation
    check_obfuscation: true                # Detect encoded payloads
    publisher_min_age_days: 30             # Minimum account age
  injection_defense:
    enabled: true
    confidence_threshold: 0.7              # Detection confidence cutoff
    use_nemo_guardrails: true              # Use NeMo jailbreak model
    sanitize_content: true                 # Auto-strip injection payloads
  hardening_auditor:
    enabled: true
    detect_drift: true                     # Alert on config changes between audits
    min_token_length: 32                   # Minimum auth token strength
  credential_guard:
    enabled: true
    redact_outbound: true                  # Auto-redact secrets in outbound data
    exfiltration_detection: true           # Detect credential-read-then-network patterns
    bulk_read_threshold: 5                 # Memory files read before alerting

For a full walkthrough of each workstream, see the OpenClaw Sentinel feature page.