One Assistant. Many Specialists.
Your main agent can spawn sub-agents — specialized workers for specific tasks. Think of it as building your own AI team.
🤖 The Main Agent
When you deploy BlitzClaw, you get a main agent. This is who you talk to on Telegram. It handles your day-to-day requests, remembers your preferences, and manages your workspace.
The main agent:
- • Maintains long-term memory across sessions
- • Has full access to tools (web search, files, browser)
- • Responds to your messages directly
- • Can delegate work to sub-agents
👥 What Are Sub-Agents?
Sub-agents are isolated workers that run specific tasks in the background. They have their own context, can use different models, and report back when done.
🔬 Research Agent
"Deep dive into competitor pricing and summarize findings"
✍️ Writing Agent
"Draft 5 variations of this email, different tones"
📊 Analysis Agent
"Review this data and find anomalies"
🔄 Monitor Agent
"Check this website daily and alert me to changes"
💡 Why Use Sub-Agents?
Parallel Work
Main agent stays responsive while sub-agents work in the background. Start a research task and keep chatting.
Focused Context
Sub-agents start fresh without your chat history. Better focus, no confusion from unrelated conversations.
Cost Control
Use cheaper models (Haiku) for routine tasks. Save Opus for complex reasoning.
Isolation
Sub-agents can't access your main memory. Good for sensitive or experimental tasks.
🛠️ Creating Sub-Agents
Just ask your main agent. It knows how to delegate.
Tips:
- • Be specific about what you want back
- • Mention if you want a specific model ("use Haiku for this")
- • Sub-agents auto-announce results to your chat
⏰ Scheduled Tasks
Sub-agents can run on a schedule. Set up recurring tasks that work while you sleep.
The agent sets up cron jobs that spawn sub-agents at the scheduled times. Results come to your Telegram automatically.
✨ SOUL.md — The Agent's Personality
Every agent has a SOUL.md file — instructions that define who it is. This is where personality, tone, and operating principles live.
# SOUL.md
You are a direct, no-nonsense assistant.
Skip pleasantries. Get to the point.
When uncertain, ask one clarifying question.
Never apologize for being helpful.
You can edit SOUL.md through your dashboard or by asking your agent to update it. Changes take effect immediately.
🎭 Presets & Specialist Personalities
Don't start from scratch. Use presets — pre-configured personalities and skills for common use cases.
📝 Content Writer
SEO-aware, adapts to brand voice, handles blogs to social posts
💻 Code Reviewer
Security-focused, knows common patterns, suggests improvements
📊 Data Analyst
Finds patterns, creates summaries, explains insights clearly
🎯 Sales Assistant
Drafts outreach, researches prospects, tracks follow-ups
📚 Research Specialist
Deep dives, synthesizes sources, cites everything
🤝 Meeting Assistant
Takes notes, extracts action items, sends summaries
Browse presets at agentpresets.ai
Community-curated agent configurations. Available now for self-hosted OpenClaw instances, coming soon to BlitzClaw with one-click install.
👥 Teams of Specialists
Combine specialists into teams that work together on complex workflows.
🚀 Product Launch Team
Research Agent → Content Writer → Social Media Manager
"Research competitors, write launch copy, schedule social posts"
📈 Growth Team
Monitor Agent → Analyst → Sales Assistant
"Track mentions, analyze sentiment, draft outreach to engaged users"
📝 Content Pipeline
Research Specialist → Content Writer → Editor Agent
"Research topic, write draft, review and polish"
🔮 The Future: Full Autonomy
Today, you orchestrate. Tomorrow, agents coordinate themselves.
Inter-agent communication is coming. Your role shifts from operator to director — setting vision and reviewing outcomes while your AI team handles execution.
Start with one agent. Scale to a team.
Your main agent is ready. Sub-agents are one message away.