Architecture

Clawdbot

What is Clawdbot?

Clawdbot is an open-source AI agent framework that runs locally on your machine, giving you a personal AI assistant that can interact with your tools, files, and services. Unlike cloud-only AI assistants, Clawdbot operates from your own computer with full access to your local environment—reading files, executing commands, browsing the web, and connecting to your services.

Think of it as having an AI employee that sits at your workstation, with access to the same tools you use, available 24/7.

Key characteristics:

  • Local-first: Runs on your machine, keeps data under your control
  • Tool-integrated: Connects to files, browsers, terminals, APIs, and external services
  • Skill-based: Modular capabilities that extend what the agent can do
  • Memory-enabled: Maintains context across sessions through workspace files
  • Multi-channel: Communicate via terminal, Discord, WhatsApp, or other integrations

How does Clawdbot work?

Clawdbot operates as an always-available agent with access to your local environment:

1. Workspace as memory Your Clawdbot lives in a workspace directory containing:

  • SOUL.md: The agent's personality and instructions
  • USER.md: Information about you and your preferences
  • MEMORY.md: Long-term curated memories
  • memory/: Daily logs and working notes
  • skills/: Modular capability definitions

The agent reads these files each session, maintaining continuity without cloud storage.

2. Tool access Clawdbot has access to your machine through integrated tools:

  • File operations: Read, write, edit files in your workspace
  • Terminal execution: Run shell commands, scripts, development tools
  • Web browsing: Control browsers for research and automation
  • Communication: Send messages, emails, manage calendars

3. Skills system Skills are modular capabilities that extend Clawdbot's abilities:

  • Camera and screen capture
  • Voice synthesis (TTS)
  • Calendar management
  • Email integration
  • Custom skills you define

Each skill includes instructions and tool configurations the agent can invoke.

4. Multi-channel communication Interact with Clawdbot through:

  • Direct terminal conversations
  • WhatsApp messages
  • Discord channels
  • API calls for automation

Same agent, different interfaces for different contexts.

5. Heartbeat and proactivity Clawdbot can run periodic heartbeats—checking in on emails, calendar events, notifications—and proactively reach out when something needs attention.

Key features

Local execution Commands run on your machine. File edits happen directly. Browser automation uses your actual browser. This enables workflows impossible with cloud-only AI.

Privacy by default Your data stays on your machine. Conversations, files, and context don't leave your control unless you explicitly integrate external services.

Persistent memory Unlike stateless chat sessions, Clawdbot maintains memory through workspace files:

  • Daily logs track what happened
  • MEMORY.md stores curated long-term knowledge
  • Project context persists across sessions

Multi-agent capable Spawn sub-agents for specific tasks. The main agent can delegate work to specialized sub-agents that complete tasks and report back.

MCP integration Connect to Model Context Protocol servers for standardized tool access. Use existing MCP servers or build custom ones for your systems.

Customizable personality SOUL.md defines who your agent is. Adjust tone, capabilities, boundaries, and behavior to match your needs.

Use cases

Personal productivity

  • Manage calendar and schedule
  • Process and respond to emails
  • Research topics and summarize findings
  • Organize files and projects

Development assistant

  • Navigate codebases and explain architecture
  • Write, test, and commit code
  • Run builds and deployments
  • Debug issues with local environment access

Content creation

  • Draft documents and communications
  • Generate and edit images
  • Create presentations and reports
  • Manage publishing workflows

Automation hub

  • Monitor systems and alert on issues
  • Execute scheduled tasks
  • Coordinate between services
  • Handle routine workflows

Research and learning

  • Search and synthesize information
  • Maintain knowledge bases
  • Track topics and news
  • Generate summaries and analyses

Setting up Clawdbot

Installation

npm install -g clawdbot

Initialize a workspace

clawdbot init ~/my-workspace

Configure your agent Edit the generated files:

  • SOUL.md: Define personality and capabilities
  • USER.md: Describe yourself and preferences
  • Add skills for additional capabilities

Start interacting

clawdbot chat

Or connect via configured channels (Discord, WhatsApp, etc.)

Clawdbot architecture

Gateway The central daemon that manages:

  • Agent sessions and state
  • Tool execution
  • Channel connections
  • Sub-agent coordination

Agents LLM-powered agents that reason about tasks:

  • Read context from workspace
  • Decide on tool usage
  • Generate responses and actions
  • Maintain conversation state

Channels Input/output interfaces:

  • Terminal (direct interaction)
  • Discord (bot integration)
  • WhatsApp (messaging)
  • API (programmatic access)

Skills Modular capability packages:

  • Define available tools
  • Include usage instructions
  • Configure permissions
  • Package for sharing

Best practices

Start with clear boundaries Define in SOUL.md what the agent should and shouldn't do. Be explicit about sensitive operations requiring confirmation.

Build memory incrementally Start with minimal context. Add to MEMORY.md and workspace files as you work together. The agent learns what's relevant.

Use skills for specialization Don't overload the base agent. Add skills for specific domains—one for email, one for development, one for calendar.

Review actions on sensitive operations Configure confirmation requirements for actions like sending emails, deleting files, or external communications.

Maintain workspace hygiene Periodically review and prune memory files. Remove outdated context. Keep instructions focused.

Customize for your workflow Clawdbot is a framework, not a finished product. Adapt it to how you work rather than changing your workflow to match defaults.

Clawdbot vs other AI assistants

vs ChatGPT/Claude web interfaces Web interfaces are stateless—every conversation starts fresh. Clawdbot maintains context and has local tool access. No copy-paste workflow.

vs Copilot/Cursor These are IDE-specific. Clawdbot works across your entire environment—files, terminal, browser, communications.

vs AutoGPT/AgentGPT Early autonomous agents ran in loops without good grounding. Clawdbot balances autonomy with workspace-based memory and human touchpoints.

vs Custom LangChain agents Clawdbot provides the infrastructure so you don't build from scratch. Workspace conventions, channel integrations, and skill system are ready to use.

The vision: AI as collaborator

Clawdbot represents a shift in how we interact with AI—from asking questions to working together. The AI becomes a persistent collaborator that:

  • Knows your context and preferences
  • Has access to your tools and systems
  • Remembers what you've done together
  • Proactively helps without constant prompting

This isn't artificial general intelligence. It's a focused, capable assistant that amplifies what you can accomplish.


Interested in building your own AI agent? Chipp makes it easy to create custom AI assistants with your knowledge, custom actions, and integrations—no coding required.

Build AI agents with Chipp

Create custom AI agents with knowledge, actions, and integrations—no coding required.

Learn more