OpenClaw Kya Hai?

Short answer

OpenClaw ek self-hosted agent runtime hai. Aap isko ek server (usually ek VPS) par install karte hain, aur yeh AI models (jaise Claude ya GPT) ko messaging channels (jaise Telegram ya WhatsApp) aur tools (jaise file access ya web search) se connect karta hai. Woh central process jo sab kuch coordinate karta hai, usko Gateway kahte hain.

Agar aap yahan isliye hain kyunki aap OpenClaw setup karna chahte hain, toh neeche diye gaye moving parts ko samajhne se shuru karein, aur fir recommended beginner path ko follow karein.

Who this is for

Yeh guide kisi aise vyakti ke liye hai jisne OpenClaw ke baare mein suna toh hai lekin abhi tak yeh clear nahi hai ki:

• Gateway aakhir karta kya hai
• Channels, models, aur tools kaise connect hote hain
• “Self-hosted” ka actual practice mein kya matlab hai
• Security optional kyun nahi hai
• Ek aise agent ko chalane ke kya risks hain jiske paas real tool access ho

Char main components

1. Gateway

Gateway ek core server process hai. Yeh ek Node.js application hai jo aapke VPS par chalta hai aur by default port 18789 par listen karta hai.

Gateway:

• Connected channels se aane wale incoming messages receive karta hai
• Messages ko configured AI model tak route karta hai
• Model dwara request kiye gaye tool calls ko execute karta hai
• Monitoring aur configuration ke liye web dashboard serve karta hai
• Authentication, channel state, aur session persistence ko manage karta hai

Agar Gateway ruk jata hai, toh sab kuch ruk jata hai — na koi messages process hote hain, na koi tools execute hote hain, aur dashboard unreachable ho jata hai.

2. Channels

Channel ek messaging platform hota hai jo Gateway se connect hota hai. OpenClaw multiple channels ko support karta hai:

• Telegram — Indian builders ke liye sabse common. Yeh BotFather tokens use karta hai aur allowlist-based access control ko support karta hai.
• WhatsApp — iske liye phone number aur QR-code session ki zaroorat hoti hai. Personal contact access ki wajah se isme additional privacy risks hote hain.
• Web dashboard — ek browser-based interface jise Gateway khud serve karta hai.
• Other channels — aapke configuration ke hisaab se Discord, Slack, Matrix, aur bhi bahut kuch.

Har channel Gateway ke saath authenticate karne ke liye ek token ya session ka use karta hai. Agar yeh tokens leak ho jate hain, toh koi dusra aapke bot ko impersonate kar sakta hai ya aapke messages padh sakta hai.

3. Models

Models AI providers hote hain jo agent ki reasoning ko power dete hain. OpenClaw in providers se connect hota hai:

• Anthropic (Claude)
• OpenAI (GPT)
• Local models (Ollama ya similar ke through)

Har provider ko ek API key ki zaroorat hoti hai jo aapke server ki environment file (.env) mein store hoti hai. In keys ka bill aata hai — agar kisi ko aapke Gateway aur aapke API key ka access mil jata hai, toh woh aapke credits consume kar sakte hain.

4. Tools

Tools aisi capabilities hain jinka use agent kar sakta hai. Examples mein shamil hain:

• File system access (aapke server par files ko read, write, delete karna)
• Shell command execution
• Web search
• External services ke liye API calls
• Database queries

Tools sabse high-risk component hain. Ek misconfigured tool permission agent ko (ya usko message bhej pane wale kisi bhi vyakti ko) aapke server par arbitrary commands execute karne ki permission de sakti hai.

Yeh pieces connect kaise hote hain

┌─────────────┐     ┌─────────────┐     ┌──────────────┐
│  Telegram    │────▶│             │────▶│  AI Model    │
│  WhatsApp    │     │   Gateway   │     │  (Claude,    │
│  Dashboard   │◀────│  :18789     │◀────│   GPT, etc.) │
└─────────────┘     │             │     └──────────────┘
                    │             │
                    │             │────▶ Tools (files,
                    │             │◀──── shell, APIs)
                    └─────────────┘

1. Ek channel se message aata hai (jaise, Telegram).
2. Gateway isko receive karta hai aur configured AI model ko bhejta hai.
3. Model respond karta hai, aur ho sakta hai tool calls request kare.
4. Gateway tools ko execute karta hai aur results model ko wapas deta hai.
5. Final response channel ke through wapas bhej diya jata hai.

”Self-hosted” ka matlab kya hai

OpenClaw aapke server par chalta hai. Ek SaaS product ke opposite:

• Aap VPS, iske operating system, iske firewall, aur iske updates ko manage karte hain.
• Aap API keys, bot tokens, aur session data ko store karte hain.
• Aap decide karte hain ki Gateway ko kaun access kar sakta hai aur agent kin tools ka use kar sakta hai.
• Security ki responsibility aapki hai — aapko protect karne ke liye yahan koi hosted backend nahi hai.

Yeh aapko full control aur full responsibility deta hai.

Security posture summary

Component	Risk if compromised	Key protection
Gateway process	Full system access	Localhost par bind karein, SSH tunnel ya Tailscale ka use karein
Channel tokens	Bot impersonation, message interception	.env mein chmod 600 ke saath store karein, Git par kabhi commit na karein
API keys	Financial loss (credit consumption)	Tokens ke jaisa hi — strict permissions ke saath .env
Tool permissions	Server par arbitrary command execution	Public channels connect karne se pehle review aur restrict karein
Dashboard	Config aur log exposure	Sirf SSH tunnel ke through access karein — kabhi publicly expose na karein

Common misconceptions (Galatfehmiyan)

“OpenClaw sirf ek chatbot hai.” — Nahi. Yeh tool access ke saath ek agent hai. Ek chatbot messages ka reply karta hai. Ek agent files read kar sakta hai, commands execute kar sakta hai, API calls kar sakta hai, aur aapke server ko modify kar sakta hai.

“Mujhe dashboard use karne ke liye port 18789 open karna padega.” — Nahi. Private access ke liye ek SSH tunnel ya Tailscale ka use karein. Port open karna aapke Gateway ko internet par expose kar deta hai.

“ufw allow 18789 run karna ek quick fix hai.” — Yeh aapke pure AI infrastructure ko public internet scanners ko expose karne ka ek quick tarika hai. Safe alternatives ke liye port 18789 guide dekhein.

“Main apna personal WhatsApp number safely use kar sakta hu.” — Personal number use karne ka matlab hai ki agent aapke private contacts ke messages process kar sakta hai. Iske bajaye ek dedicated WhatsApp number ka use karein.

Kya NAHI karna hai

• OpenClaw ko root user ke roop mein install ya run mat karein.
• Jab tak aap TLS, auth, aur reverse proxy hardening nahi samajhte, Gateway ko 0.0.0.0 par bind mat karein.
• Public forums, screenshots, ya GitHub issues mein bot tokens, API keys, ya session data share mat karein.
• Open DM access wale bot par shell execution tools enable mat karein.
• Channels connect karne se pehle VPS security checklist ko skip mat karein.

Next steps

1. Recommended beginner path ke liye Yahan Se Shuru Karo page padhein.
2. Agar aapke paas pehle se VPS hai, toh Fresh Ubuntu VPS Setup par jayein.
3. Agar aap port 18789 ko samajhna chahte hain, toh Gateway port guide padhein.
4. Agar aap terms check karna chahte hain, toh Glossary browse karein.

Madad kab maangein

Agar aap sure nahi hain ki aapka setup safe hai ya nahi, ya agar aap aise errors dekh rahe hain jo aapko samajh nahi aa rahe, toh Setup Help service ka use karein. Hum sirf redacted logs aur high-level descriptions ka use karke aapke architecture ka review kar sakte hain — hum aapse kabhi bhi raw API keys, bot tokens, WhatsApp session data, ya root passwords nahi maangenge.