Welcome

AI Governance Gateway & MCP Firewall — Policy Enforcement & Financial Accountability for AI Agents.

Last Updated: March 09, 2026 (Narrative Pivot & Architecture Clarity) Version: v2.0.4

ABS Core is a high-performance Governance Gateway and MCP Firewall designed for autonomous AI agents. It acts as a deterministic security perimeter, evaluating every intent against cryptographically versioned policies before execution.

2-Layer Architecture: Determinism & Auditability

The system separates synchronous enforcement from asynchronous auditing to maintain ultra-low latency without sacrificing compliance depth.

Layer 1: Enforcement (Sync)

Real-time blocking at the Edge (< 5ms).

WASM Policy Engine: Low-overhead, Rust-powered rule evaluation.
MCP Firewall: Deep inspection of Model Context Protocol tool calls.
Financial Guardrails: Sync checks on budget limits and staking (Bond).
Quorum: Forced human-in-the-loop for high-risk operations.

Layer 2: Compliance (Async)

Irrefutable Evidence & Reporting.

Immutable Ledger: Cryptographic audit trail of all agent actions.
Identity attribution (OID): Verified origin and intent tracking.
Policy Transparency: Public (or private) evidence of compliance.
Behavioral Telemetry: Async analysis of agentic patterns.

The Core Flow: From Intent to Enforcement

How ABS Core prevents unauthorized actions in milliseconds:

1. Interception

The Gateway intercepts the API call (OpenAI, Anthropic, or MCP) containing the agent's intent.

2. Fast Evaluation

The WASM Engine executes the active policy bundle. It checks permission, risk scores, and financial limits in < 5ms.

3. Deterministic Decision

If the action is unauthorized or exceeds safety limits, the Gateway returns a 403 Forbidden to the agent before the external tool is actually called.

4. Persistent Record

The decision (Allow/Block) and its evidence are sent to the Ledger asynchronously for auditing.

The 30-Second Hook: Secure Your Agents

The ABS Core BRIDGE stops rogue agents from dropping production tables or exposing secrets instantly.

Three Ways to Deploy

Gateway Mode (Proxy): Just set the LLM's base_url to our gateway.
Sidecar Mode: Run the WASM engine locally for zero-latency enforcement on-prem.
SDK Mode: Embed governance logic natively into your Python or Node.js application.

[!WARNING] Proxy Limitations: When running in "Gateway Mode", ABS Core intercepts Layer 7 (API) calls routed through it. It does not control the host runtime (CPU, memory, filesystem) and cannot prevent network calls that are coded to bypass the proxy. For full host isolation, use Sidecar Mode with restricted container environments.

Example: The 30-Second Gateway Setup

Point your existing LLM client to the ABS Gateway.

from openai import OpenAI

client = OpenAI(
    api_key="your-openai-key",
    base_url="https://api.abscore.app/v1" # Traffic is now routed through the Gateway
)
# Your autonomous tool calls are now governed by Zero-Latency policies.

Ecosystem & Adapters

ABS Core provides unified governance for leading frameworks:

LangChain Adapter: Plug-and-play middleware.
Crew AI Extension: Consensus kernel for multi-agent swarms.
Vercel AI SDK: Integration via custom model providers.

Technical Glossary

OID (Agent Identity)

Unique cryptographic identifier for agents, ensuring non-repudiation of actions.

WASM Policy Engine

High-security Rust-based execution layer for policy validation in < 5ms.

Bond / Slash

Financial accountability mechanism where agents stake collateral (Bond) which is "Slashed" upon policy violation.

Quorum

Consensus mechanism requiring M-of-N human or supervisor approvals for critical operations.

Immutable Ledger

A tamper-proof SHA-256 history of all audited agent interactions.

Policy as Code (DSL)

Declarative language to define complex governance rules for agent behavior.

Legal & Trust

Ready to deploy? Visit our Quickstart Guide to install the SDK.