EnforceCore

The runtime enforcement layer for agentic AI systems.

EnforceCore provides mandatory, policy-driven enforcement at every external call boundary — tool calls, API calls, file access, network access — so that policy violations become structurally impossible, not just discouraged.

---

Why EnforceCore?

Most agent safety solutions operate at the prompt level — they ask the LLM to be safe. This can be bypassed, jailbroken, or simply ignored.

EnforceCore operates at the runtime boundary — the moment before a tool or API is actually called. At this layer, enforcement is mandatory, not advisory. If a call violates policy, it doesn't execute. Period.

from enforcecore import enforce

@enforce(policy="policies/strict.yaml")
async def search_web(query: str) -> str:
    """This call is now policy-enforced, PII-redacted, and audit-logged."""
    return await api.search(query)

What It Does

Component	Purpose
Policy Engine	Declarative YAML policies — define what's allowed, denied, and how violations are handled
Enforcer	Intercepts every external call and enforces policies before execution
Redactor	Real-time PII detection and redaction on inputs and outputs
Auditor	Tamper-proof Merkle-tree audit trail for every enforced call
Guard	Resource limits (time, memory, cost) with hard kill switch

Quick Start

Install

pip install enforcecore

Define a Policy

# policies/my_policy.yaml
name: "my-agent-policy"
version: "1.0"

rules:
  allowed_tools:
    - "search_web"
    - "read_file"
    - "calculator"

  pii_redaction:
    enabled: true
    categories: [email, phone, ssn, credit_card]

  resource_limits:
    max_call_duration_seconds: 30
    max_cost_usd: 5.00

on_violation: "block"

Protect a Tool Call

from enforcecore import enforce

@enforce(policy="policies/my_policy.yaml")
async def search_web(query: str) -> str:
    return await external_api.search(query)

# ✅ Allowed tool + PII redacted from query + audit entry recorded
result = await search_web("find info about john@example.com")

# ❌ Blocked — tool not in allowed list
@enforce(policy="policies/my_policy.yaml")
async def send_email(to: str, body: str):
    await smtp.send(to, body)  # Never executes

Verify the Audit Trail

enforcecore verify audit.jsonl
# ✅ Trail integrity verified: 42 entries, chain intact, root hash: sha256:abc...

Framework Integrations

EnforceCore works with any Python-based agent system:

• LangGraph — example
• CrewAI — example
• AutoGen — example
• Plain Python — just use the @enforce() decorator

Key Design Decisions

• Async-native — supports both sync and async from day one
• Cross-platform — core enforcement works on Linux, macOS, and Windows
• Fail closed — if enforcement logic fails, the call is blocked (never fails open)
• Zero framework lock-in — no hard dependencies on any agent framework
• Honest benchmarks — real overhead numbers, not marketing claims

Performance

Component	Typical Overhead
Policy evaluation	< 1ms
PII redaction	5-15ms
Audit entry	< 1ms
Total (typical)	8-20ms

This is negligible compared to tool call latency (100ms-10s for API calls).

Documentation

Document	Description
Vision	Why EnforceCore exists and where it's going
Architecture	Technical architecture and design decisions
Roadmap	v1.0.x incremental release plan
API Design	Public API surface and usage patterns
Tech Stack	Technology choices and rationale
Developer Guide	Setup, coding standards, and workflow
Contributing	How to contribute

Roadmap (v1.0.x)

Release	Focus	Status
v1.0.0	Core Enforcer + Policy Engine	🔨 In Progress
v1.0.1	PII Redactor	Planned
v1.0.2	Merkle Auditor	Planned
v1.0.3	Resource Guard + KillSwitch	Planned
v1.0.4	Framework Integrations	Planned
v1.0.5	Evaluation Suite	Planned
v1.0.6	Hardening + Polish	Planned

See roadmap.md for detailed scope of each release.

License

Apache 2.0 — use it freely in open-source and commercial projects.

Contributing

We welcome contributions, especially:

• New adversarial evaluation scenarios
• Framework integration adapters
• PII entity recognizers
• Performance benchmarks

See contributing.md for guidelines.

---

EnforceCore is developed by akios-ai and was born from the enforcement engine inside AKIOS, a production runtime for secure multi-agent systems.