Aegis is building security for the agentic eraV1.0 Release Candidate → GitHub

How it works

Observe. Detect. Understand. Decide. Enforce.

Aegis becomes the control layer between your agents and the real world. Here is exactly what happens on every action.

  1. The agent acts

    An autonomous agent decides to call a tool — a database, a browser, a shell, an external API. It cannot execute yet.

  2. Aegis intercepts

    Through the Python SDK, REST gateway or MCP Security Gateway, the action is submitted to Aegis before execution.

  3. Identity is verified

    The Permission Engine checks the agent's registration, tool bindings and user delegation. The Trust Engine scores the agent and session trust class.

  4. Policies decide

    The deterministic Policy Engine evaluates active rules with strict precedence. No matching policy means deny — the default is closed.

  5. Risk and threats are scored

    An explainable Risk Engine produces a 0–100 score from operation sensitivity, environment and trust. Threat detectors scan payloads for injection and dangerous patterns.

  6. The decision is enforced

    ALLOW, BLOCK or REVIEW — with every reason recorded. REVIEW creates a human approval request that pauses the action until a person decides.

  7. Everything is audited

    Every evaluation lands in an append-only audit trail with redacted parameters, full provenance and correlation IDs for end-to-end reconstruction.

The evaluation pipeline, in code order.

This is the real sequence implemented in the Aegis evaluator: permission and trust run first and can short- circuit to BLOCK; policy sets the base decision; risk and threat thresholds can escalate it; human approvals resolve REVIEW outcomes.

Read the docs
  1. ACTION

    Agent intent is intercepted

  2. CONTEXT

    Agent, tool and session verified

  3. THREAT ANALYSIS

    Deterministic detectors run

  4. RISK SCORE

    Explainable 0–100 assessment

  5. POLICY

    Rules evaluated with precedence

  6. DECISION

    ALLOW · REVIEW · BLOCK