Your Local Automation AI Agent — Open Source, Self-Hosted, Never Raises Prices
Clawdbot - Your Digital Employee Never Sleeps · Never Leaks · Never Raises Prices
Clawdbot is your open-source, local-first AI digital workforce. One-click Docker deployment, seamlessly integrated with Ollama local models. Achieve 100% data privacy, automate browser, email, and calendar tasks. Say goodbye to cloud subscription fees, build your private AI workflow now.
3 AM. You're sleeping.
On your
terminal screen, code is scrolling rapidly:
Self-hosted deployment. Chat history, files, permissions all in your hands. Your data is your data, not training someone else's model.
Supports Telegram/WhatsApp. Command it like messaging a friend. No complex dashboards needed.
More than just chat. It can read/write files, control browsers, send PRs. This is real digital labor.
Supports Docker sandbox, local Ollama models ($0 cost). Fully customizable, fully open source. Built for builders.
Smart filtering starts at 8 PM daily: auto-archive marketing and low-priority emails, identify and flag key emails requiring human attention, generate editable reply drafts with one-click sending.
Scheduled scraping of target pages, price comparisons, and competitor dynamics. Extract key changes and compile into actionable morning reports, with keyword alerts and historical trend comparison.
Read/write local codebase, run tests and generate change previews, automatically create and submit PRs, assist with repetitive refactoring and boilerplate code.
All-day calendar sync, automatically organize meeting background and participant summaries, push key info before meetings and can send invites and meeting note drafts.
Configurable headless browser for simulated login, form filling, and screenshots. Supports rhythm simulation and UA switching to counter anti-scraping, auto-scrape and export data.
Deep integration with Home Assistant. Control lights, scenes, and security devices with natural language. Turn your home into a real Jarvis.
Clawdbot's architecture follows Unix philosophy: everything is a file. Your AI personality, skill library, and memory are all stored as Markdown. Readable, auditable, version-controllable.
├── SOUL.md # Core personality: Who is your AI? │ "You are a tech geek assistant who loves automation..." │ ├── AGENTS.md # Scenario roles: DevOps/Product/Support │ "When users ask deployment questions, switch to DevOps assistant..." │ ├── TOOLS.md # Tool inventory: Capability boundaries │ "read_file, send_email, browser_click..." │ ├── skills/ # Skill library: One SKILL.md per skill │ ├── gmail/ │ │ └── SKILL.md "How to read/send/classify emails" │ ├── browser/ │ │ └── SKILL.md "Headless browser operation guide" │ └── notion/ │ └── SKILL.md "Notion database CRUD" │ └── memory/ # Persistent memory ├── conversations.db (SQLite) └── vectors/ (RAG knowledge base)
All configs are Markdown. Open with VSCode and edit. No need to learn YAML/JSON.
Push the entire workspace/ folder to Git. Roll back your AI to any historical version.
Community shares SKILL.md files. Copy one .md file to gain new capabilities.
Clawdbot is based on MCP (Model Context Protocol) protocol. This means you can mount thousands of existing tools with one click, just like installing npm packages. No coding required, just one command.
Want to write your own skills?Just need one SKILL.md
and a Python/JS script. Clawdbot auto-parses inputs/outputs, AI handles the rest. Community already
has 200+ open-source skillsready to reuse.
The deeper your dependency, the harder the price hike.
Your workflow is built on a black box that could disappear at any moment.
OpenAI, Google, Anthropic are digging their moats deeper and deeper.
Before they lock it down completely, download your data, workflows, and knowledge base to your local hard drive.
This is everyone's first time using AI Agents' biggest fear. We understand. That's why Clawdbot runs in multi-layer security sandboxes by default.
Clawdbot runs in an isolated container, can only access
directories you explicitly mount (like ~/clawdbot_workspace).
Even if the code has
bugs, it can't touch your system's sensitive directories.
By default, high-risk operations require your secondary confirmation. It will ask you in Telegram: "Confirm this operation? [Yes/No]".
Enable read-only mounting for critical directories. Combined with Git version control, mistakes can be rolled back with one click.
# Security configuration example security: sandbox: true # Enable sandbox mode allowed_commands: # Command whitelist - ls - cat - echo blocked_paths: # Blocked paths - /etc - ~/Documents/private require_approval: # Operations requiring manual approval - rm - mv - send_email
In one sentence:AI is in the cage, you're outside. Even bugs can't delete your real files.
Let the numbers speak. Let's do the math: processing 100 long documents (5k tokens each) for summarization.
Our intelligent router automatically judges task difficulty, assigns simple tasks to free models, only calls paid APIs for complex tasks:
Like "convert filename to lowercase" → routed to Local Ollama
→ $0
Like "analyze complex financial report" → routed to GPT-4o / Claude
→ pay-as-you-go
Result:You get the intelligence of the strongest models, but only pay bargain prices.
vs ChatGPT Plus subscription
when using local model
Don't like one-click scripts? This is forcontrol freaksprepared manual deployment checklist (Manual Setup). Completely transparent, completely controllable.
Ensure Docker Daemon is running. Recommended
configuration:4 vCPU, 8GB RAM
(if running local model recommended 16GB)。Supports macOS, Linux, Windows
WSL2。
Create docker-compose.yaml.
This is yourdigital workforce ID card, defining its
brain (model) and hands/feet (tools).
version: '3.8' services: clawdbot: image: clawdbot/core:latest environment: # Choose your brain: Ollama or Cloud API - LLM_PROVIDER=ollama - OLLAMA_BASE_URL=http://host.docker.internal:11434 - TELEGRAM_TOKEN=your_bot_token - TIMEZONE=Asia/Shanghai volumes: # Data persistence: Your data stays local forever - ./workspace:/app/workspace - ./config:/app/config ports: - "18789:18789" # Gateway port restart: always
在 workspace/SOUL.md
write insystem prompt(this determines AI's personality
and behavior guidelines):
# Core Personality (Core Personality) You are a rigorous DevOps expert, skilled in using Linux commands. But before executing dangerous operations (like rm, mv, send_email) you must ask me first. ## Behavior Guidelines - Code output must include comments - Don't fabricate non-existent commands - When encountering vague requirements, first list 2-3 interpretation options for me to choose - Always use UTC+8 timezone
Run docker compose up -d.
Your
AI is now listening on Telegram and in the background,24/7 on standby。
Open Telegram, find your Bot, send /start
to start using. All config files are in ./workspace/
directory, you can edit with VSCode anytime.
Indie Hacker · YC W23
"I use Clawdbot to monitor 10 competitors' price changes with automated headless browser... Now Clawdbot + local Llama3.2, cost $0, 100x faster. Perfect for passive income."
Security Researcher · Ex-NSA
"The best part: even if it bugs out, it can't delete my real files. AI in the cage, me outside. This is how AI should be."
Product Manager · Ex-Google
"Like hiring an assistant who never clocks out, but I only pay $5/month. Can ChatGPT do this? No. It only chats."
(2026-01 data)
Active online users
Hybrid power strategy: Daily 80% use free local model (Ollama), complex 10% use cheap cloud model (GPT-4o mini), critical 5% use strong model (Claude Sonnet).
per day 50 emails, 40 are junk. Waste 182.5 hours per year.
This is the ultimate victory of laziness.
Your night shift trader. Drop alerts, whale monitoring, auto stop-loss.
# alert_rules.py
if price < ALERT_RULES[symbol]['support']:
await send_telegram(
f"🚨 {symbol} broke support level! Current: ${price}"
)
Wake up to see assets up by \$970. This is information advantage.
Your notes, diary, private thoughts. Completely localized, never goes to cloud.
"I feed all my articles, notes, and ideas to Clawdbot's local knowledge base. No third eye can see my drafts."
Ticket grabbing, flash sales, batch form filling. It disguises as "a really bored human".
Yes. If you have Docker. Mac/Linux one-click install. Windows supports WSL2.
Heavy use about \$15/month. Light use Heavy use about $15/month. Light use $0 (all local model).
Clawdbot is self-hosted. Data never leaves your hard drive unless you actively send it to cloud models.
ChatGPT can only "talk", Clawdbot can "do" (control browser, files, email). And Clawdbot won't steal your data for training.
| Feature | Clawdbot | ChatGPT |
|---|---|---|
| Positioning | 24×7 Digital Employee | Chat Assistant |
| Execution Capability | ✅ Send emails/Control browser | ❌ Can only chat |
| Data Storage | ✅ Your local hard drive | ❌ OpenAI servers |
| Ban Risk | ✅ 0% | ❌ 有 |
Our Discord Communityhas 2,300+ online geeks.
Whether you encounter Docker permission denied
errors, or Ollama connection refused
issues, usually within 15 minutesyou can get answers.
chmod -R 755 workspace to ensure read/write
permissions.
Includes:20+ exclusive scripts + complete configuration templates + 1-on-1 deployment guidance(limited to first 100)
FreeJoin Discord CommunityWorried about deployment? 2,300+ geeks online, average 15 minutes to solve any error.
You can continue paying \$20 per month to OpenAI as rent, or spend 10 minutes to build your own digital territory.
$ curl -fsSL https://clawd.bot/install.sh | bash