How Buda Works

If you think of Buda as just a chat tool, these concepts quickly get confusing:

• What is a Space?
• What is an Agent?
• How is Drive different from Session?
• Why do Messages affect context?
• Where do Channels, Slack, and WhatsApp fit in?

The simplest way to understand it: Buda is not a chat box — it's a company.

What you do in Buda is essentially organizing a company: there's an office space, employees, file cabinets, meeting rooms, meeting notes, and external communication lines.

One-Sentence Overview

If you only remember one sentence:

Space manages the org, Agent manages the employee, Drive manages long-term files, Session manages the current task, Message manages each conversation turn, Channel manages external entry points.

Company Org Chart Analogy

The core of this structure isn't technical jargon — it's management boundaries.

You need to know:

• What's shared at the company level
• What belongs to a specific Agent
• What's only valid in the current session
• What needs to be saved as a file for future reuse

Layer 1: Space Is the Company

A Space is like a company or a workspace.

A Space typically contains:

• Multiple Agents
• Team members
• Shared credits at the workspace level
• Shared file storage
• A set of permissions, subscriptions, and collaboration boundaries

If you have multiple businesses, multiple clients, or multiple teams, they shouldn't all be crammed into one Space.

Example:

Space A: Your own company
Space B: Client A's project
Space C: Client B's project

Each Space is like a separate company — people, files, Agents, budgets, and permissions should be clearly separated.

Layer 2: Agent Is the AI Employee

An Agent is the unit that actually does work in Buda.

Think of an Agent as an AI employee or specialist:

• Blog editor
• Xiaohongshu content writer
• Customer service assistant
• Finance reviewer
• Product research assistant
• Recruiting researcher

An Agent typically has:

• A name and identity
• Long-term instructions
• Its own tools
• Its own Drive access
• Multiple Sessions
• Installable Skills
• Bindable Channels

The most common beginner mistake is creating too many Agents right away.

A more stable approach:

If it's just the same AI employee working on different platforms, use multiple Sessions first. Only create a new Agent when identity, permissions, tools, and Drive all need to be isolated.

For example, blog, Xiaohongshu, LinkedIn, and video content can often start as four Sessions under one Agent, not four separate Agents.

Layer 3: Drive Is the File Cabinet

Drive is the long-term file storage. It solves "what materials will I need later."

Good things to put in Drive:

• Blog drafts
• Xiaohongshu drafts
• Meeting notes
• Client requirements
• Product specs
• SOPs
• FAQs
• Brand materials
• Historical cases
• Project assets

The key difference between Drive and Session:

So when the course emphasizes "saving to disk" (落盘), it means saving important chat content into Drive.

Layer 4: Session Is the Meeting Room

A Session is an independent conversation context.

Think of a Session as a meeting room or a temporary workbench.

One Agent can have multiple Sessions:

Agent: Content Operations Specialist
├── Session: Blog
├── Session: Xiaohongshu
├── Session: LinkedIn
└── Session: Video Content

Each Session has its own short-term memory.

Titles, paragraphs, and edits discussed in the Blog Session won't automatically appear in the Xiaohongshu Session. This prevents tasks from polluting each other.

If you try to write blog posts, Xiaohongshu content, and handle customer service all in one Session, it's like running three meetings simultaneously in the same room — everyone gets confused.

Better approach:

One task, one Session.

Layer 5: Message Is Each Statement in the Meeting

A Message is each individual message in a Session.

You say something, AI responds — these are all Messages.

They form the current Session's short-term context:

The longer a Session goes, the more Messages accumulate, and the larger the context becomes.

When context gets too long, the system may compress history into a summary. That's why AI sometimes "remembers the gist but forgets the details."

Channel Is an External Entry Point, Not a Memory System

A Channel is an external chat entry point, such as:

• Slack bot
• WhatsApp account
• Telegram Bot
• Discord channel
• Feishu (Lark) bot
• WeCom bot

A Channel's purpose: let team members or external users reach an Agent through familiar chat tools.

It is not a long-term memory system or a file cabinet.

A stable Channel is usually bound to a stable Agent and corresponding Session rules.

If you set up 10 bots in Slack, a more accurate way to think about it:

10 external entry points
≈ 10 work streams
≈ 10 stable sessions or session rules

Skill Is a Work Method

A Skill isn't another employee — it's a reusable methodology.

Examples:

• How to generate a PPT
• Workflow for rewriting Xiaohongshu content
• Tool for processing Excel files
• SOP for publishing articles
• Process for competitive research

If a task only happens once, don't rush to make it a Skill.

If a workflow keeps repeating — like "every time a blog post is done, rewrite it for Xiaohongshu, LinkedIn, and video" — consider turning it into a Skill.

Two Common Org Patterns

Pattern 1: One Agent, Multiple Sessions

Best for beginners and content production work.

Pros:

• Simple
• Unified materials
• Avoids premature complexity
• Perfect for getting started

Pattern 2: Multiple Agents, Collaborating via Space Shared Files

Best for teams and advanced workflows.

Pros:

• Clearer identities
• Easier permission isolation
• Team collaboration feels like a real company
• Suited for complex workflows

But beginners shouldn't start here. Master "one Agent, multiple Sessions" first.

When to Upgrade Your Structure

Use this decision flow:

Just for the current task?
→ Use the current Session.

Just switching platform style?
→ Create a new Session.

Next Session needs this content too?
→ Save to Drive.

This workflow keeps repeating?
→ Turn it into a Skill.

Team needs to use it in Slack/WhatsApp/Feishu?
→ Connect a Channel.

Identity, permissions, tools, and files all need isolation?
→ Create a new Agent.

Multiple Agents need to hand off files?
→ Use Space shared files.

Common Misconceptions

Misconception 1: Session = Agent

No. Agent is the employee; Session is a meeting that employee is currently in. One employee can be in many meetings, each with different context.

Misconception 2: Channel remembers everything

No. Channel is just an entry point. Long-term materials still go in Drive; rules go in Agent instructions or related files.

Misconception 3: Drive is chat history

No. Drive is the long-term file cabinet; chat history lives in Sessions. If you want chat content to be reusable later, actively save it as a file.

Misconception 4: Multiple platforms = multiple Agents

Not necessarily. Blog, Xiaohongshu, LinkedIn, and video can often use multiple Sessions under one Agent. Only split into separate Agents when they need different permissions, accounts, tools, or long-term materials.

Misconception 5: AI evolution means the model itself improves

Not really. In most cases, what evolves is the files:

• Materials in Drive
• Summaries in memory
• Rules in AGENTS.md
• Methods in Skills
• Historical drafts and asset libraries

The clearer these files are, the more the Agent behaves like a seasoned employee.

Remember This Diagram

Once you can distinguish these layers, you're no longer just "chatting with AI" — you're running a company.

Related Pages

• Agent
• Drive & Knowledge Base
• Chat Session