Skip to main content
GRANWEL
Products AEGIS ATTICUS Company Security Contact
Institutional Inquiries
PRODUCT SYSTEM • AUTONOMOUS ENGINEERING GOVERNANCE

AEGIS

by Granwel Developer Preview

Govern and secure autonomous engineering. AEGIS sits between autonomous AI coding agents and development environments, enforcing machine-verifiable policy for scope, security, risk, multi-agent consensus, and cryptographic audit.

Request Enterprise Preview Access Security Model

Kernel-Adapter Architecture

AEGIS decouples the governance engine from specific agent frameworks. OpenCode is the primary adapter, with Claude Code and Codex supported via standardized interfaces.

AEGIS Core Architecture & Execution Flow
SPEC: KERNEL-ADAPTER-V1
1. AGENT INTERFACE 2. ADAPTER LAYER 3. AEGIS KERNEL 4. GOVERNED REPO Autonomous Agent Claude / Codex / Custom Developer Intent Directives & Scopes UNGOVERNED RISK: • Unchecked file mutates • Scope escape • Silent hallucination OpenCode Adapter Runtime Native Tool Intercept IDE / CLI Plugs Codex & Claude Hook Session Telemetry Policy Engine (aegis.yaml) ADR Scope • High-Risk Gate • Fails Closed Consensus & Evidence Core Multi-Agent Review • Risk Matrix Deterministic Pre-flight Tests Cryptographic Audit Chain Append-only JSONL SHA-256 Merkle Provenance Verified Mutate Bounded File Edits Certified Commit Proof-of-Verification Audit Proof Log .aegis/audit.jsonl Tamper-Evident

Governance & Security Capabilities

Built on a deterministic fail-closed model where unverified mutating actions and unknown identities are denied by default.

01

Agent Governance & Identity

Authenticates agent sessions, binds tool invocations to cryptographically attested identities, and enforces scope boundaries to prevent lateral capability escalation.

02

Execution Control & Gating

Intercepts file system modifications, command executions, and network egress. Mutating actions require explicit policy validation against defined project boundaries.

03

Policy Enforcement (aegis.yaml)

Machine-verifiable rules requiring approved Architectural Decision Records (ADRs) for structural changes, cited research for external integrations, and risk classification.

04

Immutable Hash-Chained Audit

All runtime events, prompts, tool outputs, and decisions are written to an append-only JSONL log chained with SHA-256 cryptographic hashes for tamper evidence.

05

Multi-Agent Consensus

High-risk code changes require independent peer review and consensus verification from distinct subagents or designated human supervisors before merge.

06

Machine Evidence Verification

Autonomous task completion requires verified machine evidence: zero compiler errors, passing test suites, linting validation, clean container builds, and verified commit signatures.

AEGIS Verification CLI Specification

Runtime verification primitives

CLI: v1-preview
# Initialize AEGIS governance in target repository
$ aegis init
# Verify cryptographic hash chain integrity of append-only audit log
$ aegis audit verify
[AEGIS] Auditing .aegis/audit.jsonl: 142 records checked. Hash chain VALID. SHA-256 root verified.
# Verify completion certificate and machine evidence
$ aegis certificate verify --target build-cert.json
[AEGIS] Certificate SIGNED: Implementation (PASSED), Tests (100%), Docker (CLEAN). Merkle root valid.
DEVELOPMENT STATUS NOTE

AEGIS is currently in private developer preview under active enterprise development by Granwel. It is undergoing dogfood validation on Granwel corporate software and selective enterprise partner codebases.