Aegis 365 -- API Design & External Interfaces
Primary API -- OpenAI-Compatible Gateway
Aegis 365 exposes an endpoint fully compatible with OpenAI's Chat Completion API. Enterprises change BASE_URL and their entire AI stack is instantly protected -- zero code modification.
POST /v1/chat/completions
Request (OpenAI-compatible):
{
"model": "gpt-4",
"messages": [
{ "role": "system", "content": "You are a financial analyst." },
{ "role": "user", "content": "Analyze account 4111-1111-1111-1111 for John Smith." }
],
"temperature": 0.2,
"max_tokens": 1024
}
Internal processing pipeline:
- L0 intercepts; classifies action into taxonomy (A--H); routes to HITL if high-risk.
- L1 runs SLM semantic analysis; detects jailbreaks; applies confidence scoring.
- L2 detects PII/PHI; applies Private Semantic Map; outputs sensitivity tile.
- L3 tokenizes sensitive data; writes to Secure State Map; outputs anonymized prompt.
- L7 checks semantic cache; if hit, skip LLM.
- L5 routes to compliant LLM endpoint (OpenAI, Anthropic, Bedrock, Gemini).
- LLM response returned to L4.
- L4 re-hydrates tokens to real values (subject to user clearance).
- L6 generates ZKP audit evidence (async).
Response (OpenAI-compatible + Aegis metadata):
{
"id": "chatcmpl-aegis-7f3a",
"object": "chat.completion",
"choices": [
{ "message": { "role": "assistant", "content": "..." }, "finish_reason": "stop" }
],
"usage": { "prompt_tokens": 84, "completion_tokens": 210, "total_tokens": 294 },
"aegis_metadata": {
"session_id": 100237,
"layers_executed": ["L0","L1","L2","L3","L5","L4","L6"],
"policy_violations": [],
"approval_required": false,
"cache_hit": false
}
}
Authentication:
- Bearer Token: OAuth2/OIDC token from enterprise IdP (Azure AD, Okta).
- Extract user context from JWT claims (
sub,preferred_username,groups). - L0 validates user is active in
Usersand has L0 action permissions.
Admin API (Internal, Role-Gated)
| Endpoint | Purpose | Required Role |
|---|---|---|
GET /admin/audit-log |
Retrieve audit trail | SOC Analyst, CISO, Auditor |
POST /admin/break-glass |
Initiate dual-key Break Glass | Legal Counsel + CISO |
GET /admin/policies |
List policies | AIGO, CISO |
POST /admin/policies |
Create/update policy | AIGO (non-immutable), CISO (immutable) |
GET /admin/private-semantic-map |
View codename registry | AIGO, Dept Admin |
POST /admin/private-semantic-map |
Upload codename terms | AIGO only |
GET /admin/sessions/{id} |
View session state | CISO, AIGO (scoped) |
POST /admin/sessions/{id}/freeze |
Emergency session freeze | CISO, SOC Analyst |
GET /admin/siem-events |
Export to SIEM | SOC Analyst |
Third-Party Integration Endpoints
SIEM Push (L0--L6 Real-Time Events)
Syslog CEF format to Splunk, Azure Sentinel, Palo Alto XSOAR. Critical events sent immediately; low-priority batched every 5 min.
SOAR Bidirectional
- Inbound:
POST /soar/command-- Sentinel detects compromised user -- Aegis 365 freezes all sessions within 500ms. - Outbound: Anomaly flags (Mosaic patterns, high re-ID risk) sent to SOAR for enrichment.
IdP Sync
Continuous pull via Microsoft Graph / Okta API; runs every 5 min. Hard revocation: deleted user -- all active sessions frozen within 500ms.
LLM Provider Connectors
OpenAI, Anthropic Claude, AWS Bedrock, Google Gemini, Azure OpenAI (BYOK), private LLMs (Ollama, vLLM).