Workers

Workers are AI agents that execute tasks on your behalf. Each worker has its own name, description, memory, and working directory. You can manage workers through the web console or by chatting directly with the Bee (coordinator) in your IM platform.

Managing Workers Through Conversation

You can create, update, and delete workers by simply telling the Bee what you need:

Create a Worker

"Create a worker named Xiaoming to help with code reviews"

"I need a new worker for writing documentation, call it DocWriter"

The Bee will create the worker and confirm with you. You can specify a name, description, and initial memory in your message.

Update a Worker

"Update Xiaoming's description to focus on Go code reviews"

"Change DocWriter's memory to always write in Chinese"

Delete a Worker

"Delete the DocWriter worker"

"Remove the worker named Xiaoming and clean up its files"

List Workers

"Show me all workers"

"What workers do I have?"

Check Worker Status

"What's the status of Xiaoming?"

"Show me the system status"

Worker Properties

Each worker has the following properties:

How the Bee Routes Tasks to Workers

When you send a message, the Bee decides which worker should handle it based on these rules (in priority order):

1. Explicit mention — If you name a worker, it goes directly to that worker. E.g., "Xiaoming, review this PR"
2. Conversation continuation — Follow-up messages go to the same worker as the previous task. E.g., "Make it more concise"
3. Description matching — The Bee matches your request against worker descriptions semantically
4. Fallback — If no worker matches, the Bee will suggest creating one

Best Practices

Write Clear Descriptions

The description is the most important field for task routing. The Bee uses it to match your requests to the right worker.

Good descriptions — specific, skill-focused:

• "Helps with Go backend development, code reviews, and API design"
• "Writes and maintains project documentation in Chinese and English"
• "Monitors system health and generates daily status reports"

Poor descriptions — vague, unhelpful for routing:

• "Worker 1"
• "Does stuff"
• "General helper"

When multiple workers have overlapping descriptions, the Bee will ask you to choose. Keep descriptions distinct so routing is automatic.

Use Memory to Shape Behavior

Memory is written to the worker's system prompt and persists across all tasks. Use it to define the worker's personality, expertise, and constraints.

Domain expertise:

"You are an experienced Python developer. Always follow PEP 8 and prefer type hints."

Communication style:

"Respond concisely in bullet points. Avoid long paragraphs."

Constraints and safety:

"Never make breaking API changes without confirming with the user first."

Context and audience:

"You are helping a non-technical user. Explain everything in simple terms."

You can update memory at any time through conversation:

"Update Xiaoming's memory to include: always run tests before committing"

Set Custom Work Directories for Project Work

By default, each worker gets an isolated directory at ~/.openbee/worker/{id}. For workers that need to operate on a specific project, set the work directory to your project path:

"Create a worker named DevBot with work directory /Users/me/my-project"

This allows the worker to read, edit, and manage files in your project directly. The work directory must be an absolute path.

Use Specific Names for Easy Delegation

Give workers short, memorable names so you can easily delegate tasks:

"Xiaoming, fix the login bug"

"DocBot, update the README"

When you mention a worker by name, the Bee routes directly to that worker without any guesswork.

Create Specialized Workers

Rather than one general-purpose worker, create specialized workers for different domains. This improves:

• Routing accuracy — specific descriptions match requests better
• Response quality — focused memory produces more relevant output
• Task isolation — each worker maintains its own context and working state

Example team setup:

• DevBot — "Backend development, code reviews, bug fixes for the API server"
• DocBot — "Technical documentation, README updates, API reference writing"
• OpsBot — "System monitoring, deployment checks, infrastructure management"

CLAUDE.md Customization

Each worker has a CLAUDE.md file in its working directory that defines the worker's system prompt. This file is automatically managed by OpenBee based on the worker's memory field, but you can also edit it directly for advanced customization.

Worker Status