User & Usage Guide

Learn how to get the most out of Aigenzey Cowork. This guide walks you through everyday interactions, document analysis, structured artifact generation, multi-source deep research, and customizing agent memory and writing voice.

Getting Started & Accounts

Aigenzey Cowork supports two distinct account modes depending on your workflow:

• Personal Mode: Sign in instantly with your Google account. Ideal for personal productivity, calendar management, travel planning, personal document synthesis, and daily task automation.
• Business & Enterprise Mode: Designed for organizations and teams. Supports one-click Single Sign-On (SSO) with Microsoft Entra ID (Azure AD), Google Workspace, Okta, or corporate OIDC. Provides department workspaces (Engineering, Sales, Marketing, HR, Finance), shared organizational skills, role-based access control, team token budgets, and audit logging.

The Chat & Interaction Model

Aigenzey Cowork is powered by a single general-purpose orchestrator agent. When you submit a request, the agent triages the task into one of two operational paths:

1. Direct Execution (Simple Tasks): Quick lookups, calculations, email drafting, or single tool calls are answered immediately without unnecessary ceremony or multi-agent overhead.
2. Autonomous Multi-Step Workflows: Complex requests involving multiple tools (e.g. cross-referencing Zendesk tickets with Stripe invoices and generating an executive report) are executed continuously in a single cohesive reasoning effort.

Live Streaming & Tool Transparency

During execution, Cowork streams its reasoning trace in real time via Server-Sent Events (SSE). You can watch:

• Tool Call Badges: Exact tool names, parameters, and live status (e.g. jira.search_issues, gmail.list_messages, snowflake.run_query).
• Progress Indicators: Step-by-step milestones as the agent retrieves documents, computes summaries, or processes files.
• Token Usage Counter: Live metering of prompt and generation tokens for complete transparency.

Response Parts & Artifacts

Once the agent finishes execution, the reply is segmented into polished typed message parts:

File Attachments & Extraction

You can attach documents, data files, and media directly to the prompt bar. Cowork extracts text and structure natively:

• Word Documents (.docx, .rtf): Full paragraph, heading, and table extraction.
• Excel Spreadsheets (.xlsx, .csv): Sheet-by-sheet schema parsing, cell value extraction, and formula inspection.
• PowerPoint Decks (.pptx): Slide-by-slide text extraction, speaker notes, and bullet hierarchy.
• PDF Documents: High-fidelity text and table extraction with OCR fallback for scanned contracts and receipts.
• Images (.png, .jpg, .webp): Multimodal vision analysis for UI mockups, charts, infographics, and receipts.

"I've attached our Q3 financial report (Q3_Financials.xlsx) and the executive board slides (Q3_Board_Deck.pptx). 
Cross-check whether all EBITDA and ARR figures quoted on Slide 4 match the underlying calculations in the 'Financials' tab of the spreadsheet. Highlight any discrepancies."

Effort Modes & Deep Research

When a problem requires comprehensive investigation across multiple sources, Cowork supports an automatic Deep Research Mode.

Triggering Deep Research

Deep mode is classified deterministically from the user's prompt without added latency. Phrasing your prompt with depth intent—such as:

• "Deep dive into the latest SEC filings for Company X and analyze risk factors..."
• "Conduct a comprehensive research analysis on competitor pricing models..."
• "Thoroughly audit our AWS IAM roles against SOC2 compliance controls..."

Deep Research Execution Procedure

When Deep Research is activated for a turn:

1. Extended Model Ceiling: Model call ceiling is elevated from 75 to 200 calls, and thinking depth is set to high.
2. Parallel Subagent Strands: The orchestrator spawns parallel subagents to investigate independent data sources simultaneously.
3. Multi-Source Corroboration: Information is validated across at least two independent sources before being finalized.
4. Structured Sourced Brief: The agent compiles a structured document complete with executive summaries, methodology notes, and cited sources.

Model Selection & Settings

Aigenzey Cowork lets users and organizations choose their preferred foundation model at any time from the chat model dropdown:

• Gemini 3.7 Flash (Default): High-speed reasoning with extensive context window and native multimodal support.
• Claude 3.7 Sonnet: Exceptional reasoning, deep analytical writing, and code generation.
• OpenAI GPT-4o / o3-mini: Versatile reasoning and mathematical problem solving.
• Auto Mode: Automatically selects the optimal model target based on organizational configuration.

Memory, Knowledge & Writing Voice

Aigenzey Cowork continuously adapts to how you work through personalized memory and voice configuration:

1. Long-Term Memory

The agent notes important context over time (e.g. your role, preferred presentation styles, frequent project names, team terminology). You can view, search, and delete specific memory facts at any time from Settings → Memory.

2. Memory Privacy & Opt-Out

If you or your organization prefer completely stateless sessions, you can toggle Enable Memory off in User Settings. When disabled, the agent never extracts or stores persistent facts from your conversations.

3. Custom Writing Voice

Define your preferred communication tone in Settings → Writing Voice. You can specify traits such as:

• "Concise and executive-ready; avoid introductory fluff and jump straight to findings."
• "Technical, precise, and structured with clear code diffs and architectural justifications."
• "Customer-facing, empathetic, and professional for ticket drafting."