Quickstart

pip install agentshield-fw

import agentshield

shield = agentshield.Shield()

@shield.protect
def execute_sql(query: str) -> str:
    return db.execute(query)

# Agent calls execute_sql("DROP TABLE users")
# -> Blocked by AgentShield: Destructive SQL detected (ASI02)

That's it. Your agent is now protected with 39 built-in safety rules.

Why AgentShield?

AI agents are production infrastructure. They execute code, manage databases, and call APIs autonomously. But:

• 88% of organizations have had agent security incidents (Gravitee 2026 Report)
• 68% cannot distinguish human from AI agent activity (CSA Survey)
• EU AI Act high-risk obligations take effect August 2026
• Every major framework is permissive by default — if a tool is registered, the agent can call it

AgentShield fixes this in 3 lines of code. Zero dependencies in the core. Sub-millisecond latency. Full OWASP coverage.

Features

Supported Frameworks

OWASP Top 10 for Agentic Applications

AgentShield maps every rule to the OWASP Top 10 for Agentic Applications.

Configuration

Create an agentshield.yaml in your project root:

mode: enforce        # enforce | monitor | dry-run

log_file: shield.jsonl

rules:
  destructive_sql:
    enabled: true
    action: deny
  credential_leak:
    enabled: true
    action: deny
  rate_limiter:
    enabled: true
    max_calls: 100
    window_seconds: 60
  cost_guard:
    enabled: true
    max_cost_usd: 10.0
  scope:
    enabled: true
    allowed_tools:
      - execute_sql
      - read_file
      - search_web

hitl:
  timeout: 300
  timeout_action: deny
  channels:
    - type: slack
      webhook_url: ${SLACK_WEBHOOK_URL}

custom_rules:
  - name: block_twitter_posts
    tool_patterns: ["post_tweet", "send_tweet"]
    action: deny
    reason: "Twitter posting requires manual review"

Performance

Installation

# Core (zero dependencies)
pip install agentshield-fw

# With specific integrations
pip install "agentshield-fw[config]"       # YAML config support (PyYAML)
pip install "agentshield-fw[mcp]"          # MCP server support
pip install "agentshield-fw[langchain]"    # LangChain adapter
pip install "agentshield-fw[crewai]"       # CrewAI adapter
pip install "agentshield-fw[openai]"       # OpenAI Agents SDK
pip install "agentshield-fw[dashboard]"    # Real-time dashboard
pip install "agentshield-fw[hitl]"         # Human-in-the-loop gateway
pip install "agentshield-fw[otel]"         # OpenTelemetry export

# Everything
pip install "agentshield-fw[all]"

# Development
pip install -e ".[dev]"

Documentation

Full documentation is available at avinash-amudala.github.io/AgentShield.

• Getting Started
• Configuration Reference
• Rules Reference
• OWASP Mapping
• API Reference

Contributing

We welcome contributions! See CONTRIBUTING.md for details.

git clone https://github.com/Avinash-Amudala/AgentShield.git
cd AgentShield
pip install -e ".[dev]"
pytest

License

MIT — Copyright (c) 2026 Avinash Amudala