Chatting with your Genie

Chatting with your Genie isn't like using a regular AI chatbot. Your Genie has access to your server's filesystem, can execute commands, spin up sandboxes, and remember context through persistent files. This article covers how to get the most out of conversations.

Written By Deva

Last updated About 2 months ago


How Chat Works

When you send a message, it flows over a WebSocket connection to the OpenClaw agent running on your server. The agent:

  1. Reads your message

  2. Considers your SOUL.md personality + custom instructions

  3. Decides what tools to use (filesystem, terminal, sandbox, etc.)

  4. Executes actions on your server if needed

  5. Streams the response back in real time

This means your Genie isn't just generating text — it's taking action on your behalf.

Chat Sessions (Threads)

Conversations are organized as sessions — separate threads you can create, rename, and switch between.

  • Create a new session — Click the new chat button or use the sidebar

  • Switch between sessions — Click any thread in the sidebar

  • Rename — Click the session title to edit it

  • Delete — Remove a session from the menu

Each session maintains its own conversation context. Use separate sessions for different projects or tasks.

What Your Genie Can Do in Chat

Read and Write Files

Ask your Genie to read, create, edit, or organize files on your server:

"Read the config.yaml file and explain what each setting does"

"Create a new directory called 'reports' and write a summary of today's logs"

All file operations happen in your persistent workspace at /home/ubuntu/workspace/.

Execute Terminal Commands

Your Genie can run any shell command:

"Install PostgreSQL and create a database called 'myapp'"

"Show me what processes are using the most memory"

"Run the test suite and tell me what's failing"

Spin Up Sandboxes

For code execution that needs isolation, your Genie creates Firecracker microVM sandboxes:

  • Configurable memory (256 MB – 1024 MB)

  • Auto-destroy after a TTL (2–30 minutes)

  • Public URL for accessing the sandbox

  • Used for running untrusted code, testing, and isolated workloads

Use Integrations

If you've connected integrations (Gmail, etc.), your Genie can:

"Check my email for messages from the team this week"

"Send a summary of today's work to my manager"

Reference Your Knowledge Base

Upload documents (PDFs, code files, notes) to your Genie's knowledge base, and it references them in conversations:

"Based on the design doc I uploaded, what are the open questions?"

Custom Instructions & SOUL.md

Personality Presets

Your Genie's behavior is shaped by SOUL.md on your server. Choose a preset in Settings → Your AI:

PresetStyle

Balanced

Professional, opinionated, code-first. Default.

Expert

Terse. Assumes senior engineer. No explanations unless asked.

Mentor

Patient and educational. Explains the "why" behind decisions.

Pair Programmer

Collaborative. Thinks aloud. Asks for your input before acting.

Creative

Witty. Uses analogies and metaphors. Great for brainstorming.

Custom Rules

Add your own rules in Settings → Custom Instructions. Best practices:

  • Keep it to 5–10 focused rules

  • Be specific: "Always use TypeScript with strict mode" beats "Write good code"

  • Include project context: "I'm building a Next.js app with Prisma and PostgreSQL"

  • Define boundaries: "Never delete files without asking first"

Custom instructions apply to all sessions and persist across conversations.

LLM Selection

Your Genie isn't locked to one AI model. In Settings → Your AI:

Default Models

Use Genie's built-in model allocation (charged via karma credits).

Bring Your Own

Connect your own LLM credentials via OAuth:

  • Claude (Anthropic) — Connect your Anthropic account

  • Codex (OpenAI) — Connect your OpenAI account

  • Gemini (Google) — Connect your Google AI account

When using your own credentials, API calls consume your provider's quota instead of Genie karma. Switch models on the fly.

Multi-Channel Access

Your Genie isn't confined to the web app. The same AI, same server, same context — accessible from anywhere:

Telegram

Chat with your Genie as a Telegram bot:

  • Text messages and voice messages

  • Full parity with browser experience

  • Start a conversation on your laptop, continue on your phone

  • Set up in Settings → Channels → Telegram

Discord

Add your Genie to Discord:

  • Responds to @mentions in channels and DMs

  • Team collaboration with shared credit pool

  • Searchable conversation history

  • Permission-based access control

  • Set up in Settings → Channels → Discord

Cross-Platform Continuity

All channels connect to the same server. Files created via Telegram are visible in the web app. Commands run via Discord appear in your terminal history. Your Genie's context is unified across all interfaces.

Karma Cost Per Message

Every message consumes a small amount of karma (credits). The cost depends on:

  • Conversation length (longer context = more tokens)

  • Model used (your own credentials = no karma cost)

  • Tool usage (sandbox creation, file operations)

The cost is shown inline in the chat interface. If your balance runs low, you'll see a warning. If exhausted, you'll be prompted to purchase more credits.

Auto-Retry

If a message fails due to a network hiccup or temporary issue, Genie retries automatically with exponential backoff:

  • 1s → 2s → 4s → 8s → 16s (up to 5 attempts)

You don't need to re-send. Just wait.

Tips

  • Your Genie remembers through files, not memory. If you want it to remember something long-term, ask it to save a note to a file.

  • Use separate sessions for separate projects. Context stays cleaner.

  • Custom instructions are your superpower. A well-written SOUL.md dramatically improves response quality.

  • Telegram for quick tasks. Send a message from your phone: "Check if the server is healthy and restart nginx if needed."

  • Bring your own LLM if you have existing API credits — saves karma.