Sentinel Fleet — Fortified Enterprise Agent Platform

Sentinel Fleet — Fortified Enterprise Agent Platform

Problem

As enterprises deploy AI agents at scale, they face critical security gaps: agents lack identity, memory is ephemeral, no registry exists, runtime is unmonitored, and model outputs are unguarded.

Solution

Sentinel Fleet is a zero-trust enterprise agent platform built on Google ADK + Gemini 3.5 Flash implementing all 6 pillars of Track 3 (Fortified Enterprise Fleet):

1. Agent Registry — CRUD, versioning, discovery
2. Agent Runtime — Async tasks with Google ADK LlmAgent pattern
3. Memory Bank — Identity-scoped persistence with HMAC integrity
4. Agent Identity — Zero-trust HMAC tokens, enrollment, revocation
5. Agent Gateway — Identity→registry→policy→armor→runtime pipeline
6. Model Armor — Gemini 3.5 Flash guardrails: prompt injection, tool poisoning, PII
7. Agent Observability — OpenTelemetry spans, tamper-evident hash chains

Unique Angle

Unlike existing frameworks that bolt security on after the fact, Sentinel Fleet is built zero-trust from the ground up. Every call passes through identity→policy→armor→runtime. Memory is HMAC-signed. Traces are tamper-evident.

Tech Stack

Python, FastAPI, Google ADK, Gemini 3.5 Flash, Google Cloud (Firestore, Pub/Sub, Cloud Trace), Vercel

What's Real vs Mocked

✅ Google ADK, HMAC identity, memory bank, model armor, observability — all real 🔄 Firestore/Cloud Trace — mocked with SQLite/asyncio (swap on GCP provisioning)

Built With

• ai-agents
• cloud-run
• fastapi
• firestore
• gemini
• google-adk
• google-cloud
• hmac
• opentelemetry
• python
• vercel
• zero-trust