Agent Firewall Template

Policy configuration for AI agent and tool-calling workflows.

Use this page when

• You are deploying an AI agent that invokes external tools and need to restrict which tools it can call.
• You need a starting policy config with prompt-injection detection, tool allow/deny lists, rate limits, and audit logging.
• You want to go from zero to a running agent-firewall gateway in under five minutes with kt init --template agent-firewall.

Primary audience

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

Policy Config

pack:
  name: "agent-firewall"
  version: "0.1.0"
  enabled: true
  description: "AI agent tool control and quality enforcement"

policies:
  chain:
    - prompt-injection
    - agent-firewall
    - pii-detector
    - quality-scorer
    - audit-logger

policy:
  prompt-injection:
    threshold: 0.8
    action: "block"

  agent-firewall:
    allowed_tools:
      - "search"
      - "calculator"
      - "file_read"
    denied_tools:
      - "shell_exec"
      - "file_write"
    rate_limit:
      max_calls_per_minute: 30
      max_calls_per_session: 200
    transaction_limit:
      max_cost_per_session: 10.0

  pii-detector:
    action: "redact"

  quality-scorer:
    min_score: 0.7
    assertions:
      - type: "model-graded-closedqa"
        threshold: 0.8

  audit-logger:
    retention_days: 365

Quick Start

# Save the Policy Config example on this page as policy-config.yaml
export OPENAI_API_KEY="sk-your-openai-key"
kt policy lint --file policy-config.yaml
kt gateway run \
  --listen 0.0.0.0:41002 \
  --policy-config policy-config.yaml

Keep the provider target in policy-config.yaml and supply the credential separately through OPENAI_API_KEY.

If you prefer the seeded starter, run kt init --template agent-firewall first and then add the provider block shown in the example config before linting and running.

For AI systems

• Canonical terms: Keeptrusts, agent-firewall, policy-config.yaml, kt init --template agent-firewall, allowed_tools, blocked_tools, rate_limit, transaction_limit, quality-scorer.
• Related policy kinds: agent-firewall, prompt-injection, pii-detector, quality-scorer, audit-logger.
• Best next pages: Agent Firewall policy reference, Prompt Injection policy, Templates overview.

For engineers

• Prerequisites: Rust-based kt CLI installed, an LLM provider API key (e.g., OPENAI_API_KEY).
• Validate: kt policy lint --file policy-config.yaml should pass with zero errors.
• Test: send a request invoking a denied tool (e.g., shell_exec) and confirm a 409 block response.
• Customize: edit allowed_tools, denied_tools, and rate_limit.max_calls_per_minute to match your agent's permitted actions.

For leaders

• Deploying this template enforces tool-level access control on AI agents, reducing the risk of unintended side effects (data writes, shell execution).
• Rate limits and transaction caps provide cost containment without requiring per-request approval workflows.
• Audit logging with 365-day retention satisfies most internal-audit requirements for traceability of agent actions.

Next steps

• Agent Firewall policy reference — full field-level documentation
• Templates overview — browse all available templates
• Prompt Injection policy — tune the injection detection threshold
• Quality Benchmarking template — add response quality gates