Office Studio

Your work shouldn't just remember what you've done. It should help you get the work done.

Office Studio is a workspace where you and your AI coding agent work together across the information that drives your day — email, calendar, tasks, files, documents, and eventually other sources of work context.

Instead of building another AI assistant, Office Studio brings your work into an environment where powerful coding agents such as Claude Code and Codex already know how to search, analyze, write code, use tools, and execute multi-step tasks.

Human when judgment matters. Agent when intelligence helps. Automation when repetition doesn't.

Why Office Studio?

A surprising amount of office work is repetitive.

Every day we:

• Read and search email
• Check calendars and prepare for meetings
• Find documents and previous conversations
• Maintain task lists
• Prepare status reports
• Follow up with people
• Collect information from different systems
• Transform spreadsheets and documents
• Repeat processes we already completed last week

AI assistants can help with individual questions, but much of the context they need still lives across separate applications.

Traditional automation can handle repetitive processes, but someone first has to define the workflow, connect the systems, and anticipate every step.

Office Studio takes a different approach.

Work with an agent first. Codify what works. Automate when it repeats.

You can start with a normal request:

"What should I focus on today?"

Your agent can work across your email and calendar instead of relying only on what you paste into a chat.

Then ask:

"Prepare me for my meetings this afternoon."

Or:

"Find everything related to Project Alpha from the last month and summarize what changed."

Or:

"Every Friday I prepare the same report. Build a workflow so I don't have to keep doing this manually."

Modern coding agents can do more than call a fixed set of predefined AI functions. They can use existing tools, query structured data, write Python or SQL when necessary, execute the work, inspect the results, and improve the process.

A successful one-time task can become a reusable workflow.

Do it manually
       ↓
Do it with an agent
       ↓
Capture what worked
       ↓
Turn it into a reusable workflow
       ↓
Run it again
       ↓
Improve it
       ↓
Automate it

The goal isn't simply to have better conversations with AI.

The goal is to gradually eliminate repetitive work.

Why a Studio?

Office Studio isn't intended to replace humans with agents.

It creates a shared working environment for both.

                 OFFICE STUDIO

 Email ───────┐
 Calendar ────┤
 Tasks ───────┤
 Files ───────┤
 Documents ───┤
 Other Data ──┘
       │
       ▼
  Local Work Context
       │
       ├──────────────┐
       ▼              ▼
     HUMAN           AGENT
       │              │
       └──────┬───────┘
              ▼
       Tools & Workflows
              │
              ▼
            Results

Humans can search, inspect, edit, decide, approve, and intervene.

Agents can search, analyze, correlate information, write code, use tools, and execute multi-step work.

Automation can take over when a process becomes predictable enough that neither needs to repeat it manually.

That's what Studio means:

A workspace where humans and agents work together.

Don't Build Another Agent

Office Studio does not ship its own AI model.

It is designed to work with the coding agents you already choose to use, including Claude Code and Codex, with room for additional agents over time.

This creates an important separation:

Office Studio owns the workspace.

It provides the connectors, local data, normalization, schemas, search, tools, and instructions that make your work understandable to an agent.

Your coding agent provides the intelligence.

It understands your request, determines how to accomplish it, uses available tools, and can write code when a predefined tool isn't enough.

Your computer provides the execution environment.

The work can happen locally using the mature environment already available through VS Code, your filesystem, Python, Node.js, SQLite, command-line tools, and extensions.

Office Studio doesn't need to recreate an entire virtual computer just so an agent can work.

Why VS Code?

VS Code has become much more than a code editor.

It already provides many of the capabilities an agentic workspace needs:

• Files and folders
• Editors
• Terminals
• Python and other runtimes
• Command-line tools
• Extensions
• Source control
• Tasks
• Local execution
• Coding agents

Coding agents already understand this environment exceptionally well.

Office Studio extends that environment from:

"Help me work with my code."

to:

"Help me work with my work."

Instead of building another isolated AI desktop application and recreating its execution environment, Office Studio brings office work into an environment where agents already know how to operate.

Bring Your Own Agent

Office Studio deliberately avoids locking your workspace to a single AI provider.

Today you might prefer Claude Code.

Tomorrow you might prefer Codex.

Another agent may become better for a particular workflow.

Office Studio's job is to provide each capable agent with the same well-organized work context and tools.

Bring your own agent. Use your own computer. Keep your workflows.

The extension ships no AI model and sends no indexed content to an AI provider.

If you choose to use an external coding agent, that agent operates according to its own configuration, permissions, and privacy policies.

Local Work Context

Office Studio currently starts with Microsoft 365:

v1

• Mail
• Calendar

The extension pulls supported work data into a local SQLite database:

workspace.db

Full-text search is provided through SQLite FTS5.

Rather than forcing an agent to repeatedly scan raw emails or rediscover the structure of your data, Office Studio also generates an agent-readable map:

• AGENTS.md
• CLAUDE.md

This tells compatible coding agents how the workspace is organized and how to query it correctly.

The result is a local work context that both humans and agents can build upon.

Over time, Office Studio is designed to expand beyond mail and calendar into other everyday work context such as tasks, files, documents, contacts, notes, and additional data sources.

From Context to Applications

Office Studio owns the foundation:

Connectors
    ↓
Normalization
    ↓
Local Index
    ↓
Schemas
    ↓
Agent Map
    ↓
Tools

But it doesn't need to predict every application you'll want.

Your agent can build on top of that foundation.

Examples might include:

• Daily work planning
• Meeting preparation
• Follow-up tracking
• Project summaries
• Email analysis
• Status-report generation
• Commitment tracking
• Calendar analysis
• Document processing
• Personal knowledge search
• Reusable office workflows

Instead of shipping hundreds of rigid features, Office Studio provides a reliable foundation from which humans and agents can create the workflows they actually need.

The Long-Term Idea: Codify Your Work

The most valuable knowledge in an organization isn't only contained in documents.

It's also contained in how people get things done.

How do you prepare this report?

How do you investigate this problem?

What information do you check before this meeting?

How do you decide who needs a follow-up?

What steps happen every Monday morning?

Traditionally, much of that knowledge lives in people's heads.

Coding agents give us a new opportunity.

A person can perform a task with an agent, refine the process until it works, and then preserve that successful process as code, instructions, tools, or a reusable workflow.

Over time:

Work becomes knowledge. Knowledge becomes workflow. Workflow becomes automation.

Office Studio is designed to provide the workspace where that transition happens.

Your workspace shouldn't just contain your work. It should learn how your work gets done.

Architecture

Office Studio is currently organized into three major areas:

extension/
    VS Code extension (TypeScript, VS Code-coupled)
    Store, search, views, and agent map

data-pipeline/
    Headless TypeScript + Python
    Providers, synchronization, extraction,
    and cold-agent test suite

plan/
    Product vision, PRD, architecture,
    and implementation plans

Full design documentation lives in plan/high-level.

Development

Requires:

• Node.js ≥ 20
• VS Code ^1.102.0
• Python ≥ 3.10 for document conversion (v1a-docs)

npm ci
npm run rebuild:electron
npm run compile
npm run check:ids
npm run check:native-layout

Press F5 in VS Code to launch the Extension Development Host and verify the activity-bar view.

Then run Spike: better-sqlite3 to verify that the native module loads correctly inside the extension host.

Test & Gate

npm run rebuild:node
npm test
npm run gate:v1a

npm test runs the Tier-A deterministic oracle and Store unit tests.

npm run gate:v1a always runs Tier A and runs the Tier-B cold-agent gate when LYNX_AGENT_CLI is configured.

The v1a Thesis Gate

The initial thesis is deliberately simple:

Given only Microsoft mail + calendar and the generated agent map, can a completely fresh coding-agent session correctly answer representative work questions on its first attempt without brute-force scanning the underlying data?

A cold Claude Code or Codex session should be able to use the generated AGENTS.md, understand the available views, and query the workspace correctly.

If that works reliably, Office Studio has accomplished its first goal:

Give an agent enough structured understanding of your work to start being useful immediately.