Open Source

Catalog building blocks, examples, and reusable patterns for the runtime governance platform.

View GitHub

Open Source

The AISecOps Interceptor is the open-source runtime governance platform for agentic AI.

It is no longer just a policy gateway. The current implementation includes Runtime Governance APIs, Replay Diff Engine, Agent Identity Layer, Compliance Evidence Export, Risk Explanation Engine, MCP Policy Proxy, Local Enforcement Mode, Runtime Budgets, structured audit events, and an interactive Replay Audit UI with execution graph visualization.

Current OSS release: v1.0.0 - Replay Diff Engine + Evidence Export
Repository: github.com/viplavfauzdar/aisecops-interceptor
License: Apache 2.0

What Is Here

AISecOps Interceptor provides a framework-agnostic runtime governance layer for agentic AI systems.

It covers:

The project demonstrates the core AISecOps runtime model:

Agent → Plan Extraction → Capability Validation → Policy Enforcement → Runtime Budgets → Runtime Controls → Execute → Audit → Replay Diff → Evidence

This is not a prototype. It is the working runtime core described throughout the AISecOps architecture, threat model, and enterprise governance material.

The Repository

github.com/viplavfauzdar/aisecops-interceptor ↗

Runtime Governance Core

AISecOps Interceptor separates planning from execution.

Instead of allowing a model or agent to directly invoke tools, the runtime constructs an explicit execution plan and evaluates it before execution.

LLM / Agent → Structured Plan Extraction
AISecOps Interceptor → Policy Enforcement + Runtime Budgets + Runtime Controls
Executor → Act
Audit Layer → Replay Diff + Evidence Export

This creates deterministic governance boundaries between:

The runtime may return:

Capability-Gated Execution

Agents do not receive unrestricted tool access.

Tool execution is evaluated against explicit capability definitions before policy enforcement occurs.

Example capability patterns:

This prevents agents from escalating privileges through prompt manipulation, ambiguous runtime state, or chained tool behavior.

Runtime Provenance

AISecOps tracks where instructions originate.

Runtime events may include provenance from:

Example provenance:

{
  "source_type": "skill",
  "source_name": "untrusted_openclaw_skill",
  "trust_level": "unverified"
}

This allows runtime investigation to answer:

What action was attempted?
Why was it attempted?
Where did the instruction originate?
Was the source trusted?

Replay APIs

AISecOps exposes replay APIs for runtime reconstruction.

GET /replay
GET /replay/{trace_id}
GET /replay/{trace_id}/summary

Replay APIs reconstruct execution history from structured JSONL audit events.

Replay data includes:

These APIs power the Replay Audit UI, replay diff review, compliance evidence export, and future governance integrations.

Replay Audit UI

The Replay Audit UI provides runtime investigation workflows for AI agent executions.

Features include:

Replay Trace List

Replay summaries expose runtime decisions, event counts, provenance trust levels, and execution outcomes.

AISecOps replay trace list

Replay Timeline

Replay timelines reconstruct planning, evaluation, approvals, execution, and final governance decisions in execution order.

AISecOps replay timeline

Execution Graphs

Execution graphs visualize provenance-aware runtime flow from instruction source through execution outcome.

AISecOps execution graph

Structured Audit Events

AISecOps emits structured runtime events for replay and investigation.

Audit events include:

This makes audit events usable as replayable forensic evidence rather than passive application logs.

Swagger API Surface

The FastAPI API exposes interactive Swagger documentation for local testing and runtime demonstrations.

AISecOps Swagger API


Replay endpoints are also exposed through Swagger, including trace replay and replay summary APIs.

AISecOps Swagger API
AISecOps Swagger API

Quick Start

# create environment
python3.13 -m venv .venv
source .venv/bin/activate

# install dependencies
pip install -r requirements.txt

# run tests
pytest -q

# run API
uvicorn aisecops_interceptor.api.main:app --reload

# run dashboard
cd dashboard
npm install
npm run dev

Python 3.11 through 3.13 supported.

Policy Bundle Example

Define runtime governance policy in YAML:

rules:
  - tool_name: restart_service
    agent_name: ops_agent
    action: require_approval

  - tool_name: read_customer
    sensitivity_level: high
    action: block

  - tool_name: send_email
    action: require_approval

Load at runtime:

policy = PolicyEngine.from_yaml("policies/production.yaml")

Bundles are validated before runtime execution.

Repository Structure

aisecops_interceptor/

  core/           interceptor, policy, approval, audit,
                  context, decision, execution, events

  replay/         replay engine, trace reconstruction,
                  replay summaries

  guard/          detectors, input inspector, output inspector

  llm/            providers, pipeline, config, models

  policy/         rule engine, schema, loader

  integrations/   LangGraph adapter, OpenClaw adapter

  api/            FastAPI runtime APIs

  dashboard/      Replay Audit UI + execution graphs

Framework Integration

The interceptor integrates with any agent framework through thin adapters.

Current integrations:

flowchart LR

A[LangGraph]
B[OpenClaw]
C[CrewAI / AutoGen]
D[Custom Runtime]

A --> E[Adapter]
B --> E
C --> E
D --> E

E --> F[AISecOps Runtime Core]
F --> G[Replay + Governance Layer]

Why This Matters

AI agents are moving from chat systems to execution systems.

They can:

Traditional application logging is insufficient for understanding runtime agent behavior.

AISecOps is designed to make runtime behavior:

The goal is not only to block unsafe actions.

The goal is to reconstruct:

what happened,
why it happened,
and where the instruction originated.

What Is Coming Next

Current roadmap focus:

V

Viplav Fauzdar

Building AISecOps as a discipline and open-source runtime governance reference implementation. Focused on practical, shipped security for agentic AI systems.

Medium ↗ GitHub ↗ LinkedIn ↗