ABS Core

ABS Core is the Elite Governance and Accountability layer for autonomous AI agents.

AI agents are executing real actions in production — writing to databases, calling financial APIs, sending emails, spending money. Most teams have no way to audit what happened, prevent what shouldn't happen, or hold an agent financially accountable for violations.

ABS Core solves this with the Hexagon of Defense:

How It Works

Your AI Agent
     ↓
Magic Proxy  ←→  Policy Engine  ←→  Entropy Monitor
     ↓                  ↓
  APPROVED           BLOCKED / HELD
     ↓
External API
     ↓
 Event Recorded → Merkle Chain → Audit Trail

Every request your agent makes passes through the Magic Proxy. The proxy evaluates the request against your governance policy and runs Shannon Entropy analysis to detect obfuscated payloads. Approved requests are forwarded. Blocked requests are rejected. High-risk requests are held for human approval.

Everything is recorded in an immutable event stream. Every event is hashed and chained — giving you a cryptographic proof of your agent's entire history.

The Financial Risk Layer (Enterprise)

While ABS Core provides general safety, it introduces specialized financial accountability for high-stakes deployments.

Agents can post a bond — collateral that gets slashed when policy is violated. This transforms compliance from a checkbox into a measurable financial instrument.

This is the architecture enterprises need when AI agents are authorized to touch real money or sensitive assets.

Who Is ABS Core For

FinTech teams deploying AI agents that interact with payment APIs, trading systems, or financial data.

HealthTech teams running autonomous workflows that touch patient data or clinical decision systems.

E-commerce platforms using AI agents for pricing, inventory, and customer interaction automation.

Any engineering team that needs to prove to a compliance officer, auditor, or regulator exactly what their AI agents did — and exactly why they were allowed to do it.

1. Route via Magic Proxy (no code changes)

# Just change base_url in your agent's LLM call:
base_url = "https://api.abscore.app/proxy/openai"

# All requests are automatically governed, logged, and audited.
# No SDK install required.

2. Or ingest events directly (MCP or HTTP)

// Via MCP Tool (Cursor/IDE)
// abs_check_command({ command: "transfer", payload: { amount: 9999 } })

// Or direct HTTP POST:
POST https://api.abscore.app/v1/events
Authorization: Bearer <ABS_SECRET_KEY>
{
  "event_type": "llm.completion",
  "payload": { "model": "gpt-4", "prompt": "..." },
  "metadata": { "agent_id": "my-agent", "tenant_id": "my-org" }
}

3. Monitor in the Dashboard

Open api.abscore.app/dashboard — filter by tenant, view decisions, Sentry issues, Datadog logs, and benchmark results.

Architecture

ABS Core is built on:

• Cloudflare Workers — Edge-deployed proxy with global low latency
• Rust Sovereign Kernel — The policy enforcement engine, built for deterministic execution
• Hardware TEE (AWS Nitro Enclaves) — Tamper-proof execution environment for the clearing house
• Merkle Chain — Cryptographic audit trail for every event

Editions