Skip to content

Agent-Aegis

The lightweight, research-grade governance library for AI agents. 1 dependency (PyYAML). 24 academic papers. Two lines of code.

pip install agent-aegis and add one line. Agent-Aegis auto-instruments LangChain, CrewAI, OpenAI Agents SDK, OpenAI, Anthropic, and 6 more frameworks at runtime -- every LLM call and tool invocation passes through prompt-injection detection, PII masking, and a full audit trail.

Try it live in your browser -- no install needed.


The Fastest Way to Add AI Agent Security

import aegis
aegis.auto_instrument()

# That's it. Every AI call in your app is now governed.
# Prompt injection detection, PII masking, toxicity filtering, audit trail -- all active.

Or zero code changes -- just set an environment variable:

AEGIS_INSTRUMENT=1 python my_agent.py

Aegis detects which AI frameworks are installed and monkey-patches them at runtime, the same approach used by OpenTelemetry for observability and Sentry for error tracking. Your existing code stays untouched.

Supported Frameworks

Framework What gets patched Status
LangChain BaseChatModel.invoke/ainvoke, BaseTool.invoke/ainvoke Stable
CrewAI Crew.kickoff/kickoff_async, global BeforeToolCallHook Stable
OpenAI Agents SDK Runner.run, Runner.run_sync Stable
OpenAI API Completions.create (chat & completions) Stable
Anthropic API Messages.create Stable

Default guardrails (all deterministic, no LLM calls, sub-millisecond):

  • Prompt injection detection -- 10 attack categories, 85+ patterns, multi-language (block)
  • Toxicity detection -- harmful/abusive content (block)
  • PII detection -- 12 categories including credit cards, SSN, API keys (warn)
  • Prompt leak detection -- system prompt extraction attempts (warn)

Beyond Auto-Instrumentation

Aegis is also a full security framework with YAML policy engine, approval gates, audit trail, and compliance reporting:

Action      Policy        Approval       Execute     Audit
  |            |              |              |           |
read CRM  --> auto (low)  --> skip -------> run ------> logged
bulk edit --> approve (high) --> human y/n -> run ------> logged
delete *  --> block (critical) ------------> X --------> logged

Key Features

Feature Description
Auto-instrumentation aegis.auto_instrument() -- one line governs LangChain, CrewAI, OpenAI, Anthropic
Runtime guardrails PII detection (12 categories), prompt injection (85+ patterns), toxicity, prompt leak
YAML policies Glob matching, first-match-wins, JSON Schema for validation
Smart conditions time_after, time_before, weekdays, param_gt/lt/eq/contains/matches
4-tier risk model low / medium / high / critical with per-rule overrides
7 approval handlers CLI, Slack, Discord, Telegram, email, webhook, or custom
Audit trail SQLite + JSONL + webhook + Python logging -- auditor-ready evidence export
7 framework adapters LangChain, CrewAI, OpenAI Agents SDK, Anthropic Claude, Playwright, httpx, MCP
REST API server aegis serve policy.yaml -- govern from any language
Web dashboard Real-time security dashboard with compliance reports and anomaly detection
CLI tools aegis init, validate, simulate, audit, serve, scan, score, probe
Type-safe Full mypy --strict, py.typed marker
< 1ms overhead Policy evaluation adds minimal latency to your agent

Real-World Use Cases

Scenario How Aegis Helps
CRM Agent Read contacts freely, review updates, block mass deletes
Code Agent Read files safely, review edits, block deploys to production
Financial Agent View accounts, approve payments over threshold, block after-hours transfers
Browser Agent Navigate freely, review form fills, block JavaScript execution
Data Pipeline SELECT freely, review INSERTs to production, block DROP TABLE
Compliance Every action logged with full context for SOC2/GDPR/HIPAA evidence

Install

pip install agent-aegis

The fastest path -- auto-instrument everything:

import aegis
aegis.auto_instrument()

For manual control, install specific integrations:

pip install 'agent-aegis[langchain]'      # LangChain
pip install 'agent-aegis[crewai]'         # CrewAI
pip install 'agent-aegis[openai-agents]'  # OpenAI Agents SDK
pip install 'agent-aegis[anthropic]'      # Anthropic Claude
pip install 'agent-aegis[httpx]'          # REST APIs
pip install 'agent-aegis[playwright]'     # Browser automation
pip install 'agent-aegis[server]'         # REST API server
pip install 'agent-aegis[all]'            # Everything

Quick Example

from aegis import Action, Policy, Runtime
from aegis.adapters.httpx_adapter import HttpxExecutor

async with Runtime(
    executor=HttpxExecutor(base_url="https://api.example.com"),
    policy=Policy.from_yaml("policy.yaml"),
) as runtime:
    # Single action
    result = await runtime.run_one(Action("get", "/users"))

    # Multiple actions with plan
    plan = runtime.plan([
        Action("get", "/users"),              # auto-execute (low risk)
        Action("post", "/users", params={"json": {"name": "Alice"}}),  # approve
        Action("delete", "/users/all"),       # blocked (critical risk)
    ])
    results = await runtime.execute(plan)

Production Ready

Aspect Detail
5,035+ tests Every adapter, handler, and edge case tested
Type-safe mypy --strict with zero errors
< 1ms evaluation Policy check adds negligible overhead
Fail-safe Blocked actions never execute, period
Audit immutability Results are frozen; audit writes happen before returning
Clean patching Auto-instrumentation is fully reversible via reset(), idempotent, skip-if-missing

Roadmap

Version Status Features
0.1 Released Policy engine, 7 adapters, CLI, audit, conditions, JSON Schema
0.1.3 Released REST API, retry/rollback, dry-run, hot-reload, 7 approval handlers
0.1.7 Released Crypto audit chain, RBAC, rate limiter, regulatory mapper, anomaly detection, policy versioning
0.1.9 Released Web dashboard, autopolicy (NL to YAML), adversarial probe, multi-tenant isolation
0.2 Released WebSocket real-time streaming, interactive playground, policy editor, shields.io badge
0.4 Released aegis.init(), runtime guardrails, AGEF/AGP open governance specs
0.4.2 Released Auto-instrumentation -- aegis.auto_instrument() for LangChain, CrewAI, OpenAI Agents SDK, OpenAI, Anthropic
0.5 Released Auto-instrumentation for LiteLLM, Google GenAI, Pydantic AI, LlamaIndex, Instructor, DSPy (11 frameworks). MCP Proxy Server. aegis plan + aegis test CLI
0.6 Released 18 security fixes, IBAN PII detection, Policy CI/CD (PolicyImpactAnalyzer, PolicyTestRunner), cost governance, compliance reports, behavioral drift detection
0.6.1 Released Guardrail performance optimization — combined regex, LRU cache. 2.65ms per-call overhead for full 4-scan stack
0.7.0 Released Streaming-aware guardrail engine (StreamingGuardrailEngine), auto strategy selection (windowed vs full-buffer), Streaming Guard playground demo
0.9.0 Released Selection Governance — ActionClaim tripartite structure, ImpactVector 6D scoring, justification gap detection, SelectionAudit, commit-reveal protocol, CircuitBreaker, AegisProxy
0.9.1 Released Quality hardening — token-boundary matching, gaming resistance, thread safety, TTL, sync decorator support
1.0 2027 Distributed security, hosted SaaS, SSO/SCIM