Agents & Coworkers

In AnyCowork, we use the term Coworker to describe an AI Agent. This isn't just a marketing term - it reflects a fundamental difference in design. A simple "chatbot" answers questions; a Coworker understands context, has specialized skills, and works with you to achieve goals.

The AnyCowork Agent

At its core, an Agent in AnyCowork is defined by:

1. Identity: A name, role, and personality (System Prompt).
2. Context: Access to specific knowledge (Files, MCP Resources).
3. Capabilities: Tools it can execute (MCP Tools, Native Tools).
4. Memory: A persistent history of interactions.

Agent Types

AnyCowork supports different types of agents for different workflows:

1. Chat Agent (Standard)

The default experience. A conversational partner that can use tools when needed. Best for brainstorming, asking questions, and simple tasks.

2. Planning Agent (Coordinator)

A high-level coordinator that breaks down complex requests into a plan. It uses a Router to classify queries as simple or complex, then either responds directly (via SimpleChatAgent) or creates a step-by-step plan and executes each task sequentially.

Example: "Refactor the authentication system to use JWTs."

• Planner: "I need to: 1. Update schema, 2. Update backend handlers, 3. Update frontend client."
• Execution: Performs each step, checking for errors along the way.

Built-in Roles

AnyCowork comes with templates for common roles. You can also create your own.

👨‍💻 Senior Developer

• System Prompt: Optimized for code analysis, debugging, and architecture.
• Tools: Filesystem access, Grep search, Git commands.
• Context: 2M token window (Gemini 3 Pro) allows loading entire codebases.

✍️ Technical Writer

• System Prompt: Focused on clarity, structure, and documentation standards.
• Tools: Markdown rendering, file reading.
• Usage: "Read these source files and generate a getting-started.md guide."

🔍 Research Analyst

• System Prompt: adept at synthesis and unbiased analysis.
• Tools: Web search (via MCP), summarization.

Creating a Custom Coworker

You can create a highly specialized coworker for your specific needs.

1. Go to the Agents tab.
2. Click + New Agent.
3. Name: Give them a distinct name (e.g., "SQL Expert").
4. System Prompt: This is the most crucial part. Be specific.

   "You are a SQL optimization expert. Your goal is to analyze queries and suggest indexes. Always explain the 'why' before showing code. Assume PostgreSQL syntax."

5. Tools: Select the tools this agent needs (e.g., Database connection, but maybe not web search).

Best Practices

• One Role per Agent: Don't try to make one agent do everything. Specialized prompts perform better.
• Provide Context: Use the @ mention feature to pull in relevant files or folders into the context window.
• Iterate on Prompts: If an agent isn't behaving as expected, tweet the system prompt. Treat it like coaching a human intern.