Gemini Enterprise Architecture Guide

The Core Distinction: Two Different "Geminis"

Critical Understanding

Google uses the name "Gemini" for multiple products with fundamentally different capabilities. Confusing them will lead to failed implementations and wasted time. Understanding them as complementary systems unlocks their full power.

🔧 The "Action Engine" — Gemini for Workspace

Purpose: Execute actions within Google Workspace apps

Side panel in Drive, Docs, Sheets, Gmail
Can WRITE: Create files, folders, move items
Operates within immediate app context
Designed for direct productivity actions

drive.google.com → "Ask Gemini"

"I do things in the file system."

🧠 The "Knowledge & State Engine" — Gemini Enterprise

Purpose: Knowledge retrieval, synthesis, memory, and structured external actions

Connects to Data Stores (Drive, SharePoint, etc.) — READ
Internal Memory — READ/WRITE (core statefulness)
Assistant Actions — WRITE (Calendar, Gmail, Jira)
Designed for enterprise search and AI advisors

vertexaisearch.cloud.google.com

"I understand, remember, and act in structured systems."

The Powerful Loop: How They Work Together

Action Engine (Gemini in Drive) creates a document → Knowledge Engine (Gemini Enterprise) reads and indexes that document via Data Store → You can now ask Gemini Enterprise questions about what the other Gemini created. This makes the entire ecosystem more stateful when used together.

Stateful Agent vs. Stateless Robot

Stateless (RAG Robot)

Each conversation starts fresh. No memory of previous sessions. Only retrieves information from connected documents. Cannot learn or adapt over time. Cannot write anywhere.

Stateful (True Agent)

Remembers information across sessions. Can modify its own internal state. Can take actions in external systems. Learns and personalizes over time.

Verified: Gemini Enterprise IS Stateful — With Two Components

Gemini Enterprise's statefulness comes from two separate capabilities, not just one. Understanding this distinction is crucial.

Component 1: Internal State (Core Memory)

Gemini Enterprise has a private READ/WRITE loop with its own Internal Memory database.

Stores facts, preferences, and context about you
Persists across all sessions automatically
Enabled by default — no configuration needed
This is the "core state" that makes it an agent

Example: "Russell's favorite strategic framework is Move 37" — stored permanently.

Component 2: External State (Structured Actions)

Gemini Enterprise can WRITE to specific external systems via pre-built Assistant Actions.

Google Calendar — create events
Gmail — draft and send emails
Jira Cloud — create tickets
ServiceNow — open incidents
Microsoft Outlook — schedule meetings

Note: Google Drive is NOT in this list — that's the key limitation.

Gemini Ecosystem Architecture

You (User)

Interact with both systems

↓ ↓

🔧 Action Engine

Gemini for Workspace

(Drive Side Panel)

| |

🧠 Knowledge Engine

Gemini Enterprise

(This Platform)

WRITE to files ↓ ↓ READ/WRITE

Google Drive

Shared Data Layer

→ READ →

Internal Memory

R/W (Core State)

↓

Assistant Actions

WRITE Capability

↓

Calendar

Create Events

Gmail

Send Email

Jira

Create Tickets

Action Engine (File System Write)

Knowledge Engine (Memory + Actions)

External Systems

The Connected Loop

Action Engine writes files to Google Drive → Knowledge Engine reads those files via Data Store connection → Knowledge Engine can now answer questions about content created by Action Engine. Meanwhile, Knowledge Engine maintains its own Memory and can write to Calendar, Gmail, Jira, etc.

Detailed Capability Matrix

System	Read	Write	Connection Type	Setup Required
Internal Memory	✓ Yes	✓ Yes	Core Feature	Enabled by default
Google Drive	✓ Yes	✗ No	Data Store	Connect via Data Stores
Google Calendar	✓ Yes	✓ Yes	Assistant Action	OAuth + Client Secret
Gmail	✓ Yes	✓ Yes	Assistant Action	OAuth + Client Secret
Microsoft Outlook	✓ Yes	✓ Yes	Assistant Action	OAuth Configuration
Jira Cloud	✓ Yes	✓ Yes	Assistant Action	API Token
ServiceNow	✓ Yes	✓ Yes	Assistant Action	Instance Configuration
SharePoint	✓ Yes	✗ No	Data Store	Connect via Data Stores
Confluence	✓ Yes	✗ No	Data Store	Connect via Data Stores

Data Stores vs. Assistant Actions

Data Stores (Knowledge)

Analogy: Giving the agent a library of books to study

Read-only access to documents
Indexes content for fast search
Used for: Drive, SharePoint, Confluence, websites
Setup: Connect via Gemini Enterprise UI

"I can find and understand information in these sources."

Assistant Actions (Agency)

Analogy: Giving the agent keys to perform tasks

Read AND write access
Executes structured operations
Used for: Calendar, Gmail, Jira, ServiceNow
Setup: OAuth + API credentials required

"I can take action in these systems on your behalf."

What You Can Build: Supreme Advisor Capabilities

Strategic Knowledge Base

READ

Search and synthesize insights from all organizational documents in connected Data Stores

Persistent Context

READ WRITE

Remember strategic decisions, preferences, and context across all sessions via Memory

Meeting Scheduling

READ WRITE

Check availability and create calendar events for strategic reviews

Communication Drafting

READ WRITE

Draft and send emails to team members via Gmail integration

Task Management

READ WRITE

Create and update tickets in Jira or ServiceNow for action items

Document Creation

NOT AVAILABLE

Cannot create files in Drive. Use Gemini for Workspace side panel instead.

Common Gotchas and Confusions

"Gemini Enterprise" name collision: Google discontinued an old add-on called "Gemini Enterprise" in January 2025 and launched a completely different platform with the same name. Search results mix documentation for both.
Gems ≠ Agents: "Gems" are saved prompt personas in consumer Gemini. They have no data access, no memory, and cannot take actions. Real agents require Gemini Enterprise.
Memory ≠ Memory Bank: "Memory" is the built-in cross-session feature (no code required). "Memory Bank" is a developer feature in Vertex AI requiring the Agent Development Kit (ADK).
OAuth scope matters: Configuring OAuth doesn't guarantee write access. The specific API scope (e.g., drive vs drive.readonly) determines permissions.
Data Store ≠ Full Access: Connecting a Data Store only grants search/read capability. It does NOT enable writing, even with full OAuth configured.
Agent tools have hard limits: The google_drive_agent tool can only search. This is a fundamental design limitation, not a configuration issue.

Setup Requirements Summary

Connection Type	What You Need	Where to Configure
Google Drive (Data Store)	Workspace connection enabled, folder ID	Gemini Enterprise → Data Stores
Google Calendar (Action)	OAuth 2.0 Client ID, Client Secret	Google Cloud Console → APIs & Services
Gmail (Action)	OAuth 2.0 Client ID, Client Secret	Google Cloud Console → APIs & Services
Jira Cloud (Action)	API Token, Instance URL	Gemini Enterprise → Actions
Microsoft 365 (Data Store)	Azure AD App Registration, Tenant ID	Azure Portal + Gemini Enterprise

The Bottom Line

The Complete Picture

Gemini Enterprise is NOT a stateless RAG robot. It is a stateful agent with two write capabilities: its own Internal Memory (core state) and structured Assistant Actions (Calendar, Gmail, Jira, etc.). The only limitation is file creation in Drive — which is handled by a separate product (Gemini for Workspace). When used together, the two Geminis form a powerful, stateful ecosystem.

🔧 For File Creation & Organization

Use Gemini for Workspace (the "Action Engine" in the Drive side panel). It can create docs, sheets, slides, and folders.

🧠 For Knowledge, Memory & External Actions

Use Gemini Enterprise (the "Knowledge Engine"). It reads Data Stores, maintains Memory across sessions, and writes to Calendar/Gmail/Jira.

🔄 For Maximum Power: Use Both

Action Engine creates files → Knowledge Engine reads them via Data Store. Knowledge Engine remembers context → Informs what Action Engine should create next. The loop makes the system stateful end-to-end.