Skip to content

AegisAgent documentation

The integrity layer for AI agent actions — delivered as an integrity-anchored Agent SOC. Open, self-hostable, framework-neutral.

Make the approval trustworthy. Trust the source, not the text. Run the SOC on the proof.

Overview

AegisAgent checks an agent's proposed action before execution, binds required approval to the exact canonical action, produces verifiable evidence, and operates a deterministic SOC on that evidence.

Status: Known-agent integrity and gateway-evidence SOC capabilities largely ship today. Unknown-agent runtime force paths and multi-replica enterprise operations remain Partial. See Implementation Status.

Why This Exists

AI agents act with real credentials on real systems. Permission without source provenance, exact-action approval, and independently verifiable evidence creates false assurance. AegisAgent provides those controls at deployable action choke points.

  • Getting started

    What AegisAgent is, its components, and how the architecture fits together.

    Overview

  • Installation

    Deploy the gateway and run your first protected action in minutes.

    Install

  • Integration & connectivity

    Connect agents from anywhere — inline SDK, proxy, or agentless.

    Connect agents

  • The Agent SOC

    Detect, correlate, contain, and prove every agent action.

    SOC design

  • Agent workforce

    Govern your AI agents as a digital workforce — directory, lifecycle, fleet.

    Workforce governance

  • Action receipts

    The open, hash-chained verifiable-evidence format.

    Receipt spec

  • Qdrant integration

    Semantic audit log indexing and vector search.

    Qdrant guide

What is AegisAgent?

AI agents now take real actions across company systems. A market of gateways can already decide whether an action is allowed — that part is commodity. AegisAgent makes those decisions trustworthy and provable, and operates a SOC on the resulting evidence.

It is built on three things a generic gateway or SIEM can't copy:

  • Approval integrity — every human approval is bound to a SHA-256 hash of the exact frozen action; the SDK fails closed if a different, edited, expired, or replayed action would execute.
  • Deterministic trust-provenance gating — authorization is gated on the source trust level of the triggering content (6 levels), not a guessable text score.
  • Verifiable, hash-chained action receipts — tamper-evident evidence for SOC 2 / EU AI Act Article 14, and the spine of the Agent SOC.

How it fits together

flowchart LR
    AGENT[Agent] -->|SDK / controlled integration| GW[Gateway]
    GW --> POLICY[Cedar + exact approval]
    POLICY --> RECEIPT[Verifiable receipt]
    GW -. async evidence .-> SOC[Detect · correlate · investigate · respond]

Start with Integration & connectivity to connect your first agent, or Installation to stand up the gateway.