Most people install OpenClaw, run a few tasks, and call it done. Power users do something different — they layer hacks on top of each other until OpenClaw feels less like a tool and more like a team member who never clocks out.
This post documents 21 hacks that experienced OpenClaw users have discovered through months of daily use. Some will save you money. Some will make your agent dramatically smarter. A few are so obvious in hindsight that you'll wonder how you missed them.
Let's start where OpenClaw's real intelligence lives: memory.
Memory Hacks
OpenClaw's memory system is plain Markdown stored on disk. The agent only "remembers" what gets written to files. This architecture is both its greatest strength and the source of most beginner frustration — once you understand it, everything clicks.
Hack 1: Use Pearl for Automatic Memory Layering
Pearl is a community-built memory layer and intelligent model router that sits between your prompts and OpenClaw's LLM calls. When you install it, Pearl automatically intercepts every conversation, extracts memorable content, and writes structured memories to your workspace — without you lifting a finger.
clawhub install pearl-memory
openclaw skill enable pearl-memory
Pearl separates memory into two types: short-term daily logs (memory/YYYY-MM-DD.md) and long-term durable facts (MEMORY.md). It decides which is which using a lightweight classifier that adds only a few milliseconds of latency. The result: an agent that remembers what you discussed last Tuesday when you ask a follow-up question today.
Hack 2: Write Your Own ABOUT.md User Profile
Before Pearl existed, power users discovered a simpler trick: create a file called ABOUT.md in your workspace root and fill it with facts about yourself.
# About Me
- Name: Rahul Sharma
- Location: Bengaluru, Karnataka, India
- Timezone: IST (UTC+5:30)
- Work: Backend engineer at a fintech startup
- Prefers: Concise answers, INR currency, Indian date format (DD/MM/YYYY)
- Languages: English (primary), Hindi (conversational)
- Key tools: VS Code, PostgreSQL, Docker, GitHub
OpenClaw reads this file at the start of every session. It costs you nothing and immediately makes every response more relevant to your actual context. Think of it as the system prompt you never have to repeat.
Hack 3: Seed Manual Memory Files for Instant Context
You do not have to wait for Pearl to learn things organically. If you start a new project, create a manual memory file right away:
# Create project context file
cat > memory/project-atlas.md << 'EOF'
# Project Atlas (started March 2026)
Stack: .NET 9, PostgreSQL 16, Next.js 15
Deployed to: Azure West India
Key contacts: Priya (PM, [email protected]), Dev lead: Karthik
Sprint cadence: 2-week sprints, Monday standups
Current sprint goal: Complete authentication module by April 4
EOF
Now when you ask OpenClaw anything about Project Atlas, it already has the full context. No onboarding required.
Hack 4: Use Session Continuity With Daily Summaries
At the end of each workday, prompt OpenClaw to write a session summary:
"Write a session summary of everything we worked on today to
memory/2026-03-26-summary.md."
This single habit pays dividends for weeks. When you return after a long weekend or a holiday, you can prime the next session with:
"Read my last three session summaries and brief me on where I left off."
Fifteen seconds instead of fifteen minutes trying to remember where you were.
Hack 5: Prune Stale Memories Monthly
Memory files accumulate fast. After a month, your MEMORY.md may contain facts that are no longer true — an old tech stack, a project that ended, a preference you changed. Stale memories actively hurt quality because the agent confidently references outdated information.
Build a monthly habit:
# Add this to your OpenClaw cron (see Hack 11)
# Every first of the month:
openclaw task "Review MEMORY.md and remove any entries older than 60 days or that reference completed projects. Ask me before deleting anything."
Clean memory is better memory. Quality beats quantity every time.
Model Routing Hacks
OpenClaw is model-agnostic — it can work with Claude, GPT-4o, Gemini, Llama, Mistral, and dozens of others. Smart model routing is where significant cost savings live.
Hack 6: Route Simple Tasks to Cheap Models Automatically
Not every task needs Claude Sonnet or GPT-4o. Summarising a bullet list, reformatting a date string, or writing a one-line commit message can be done perfectly well by a smaller, cheaper model.
Edit your ~/.openclaw/config.yml:
model_routing:
default: claude-3-5-sonnet
rules:
- pattern: "summarise|reformat|convert|translate"
model: groq/llama-3.1-8b-instant
reason: "Simple transformation tasks"
- pattern: "write code|debug|architecture|review"
model: claude-3-5-sonnet
reason: "Complex reasoning tasks"
- pattern: "search|browse|lookup"
model: groq/mixtral-8x7b
reason: "Web tasks"
This alone can cut your API bill by 40-60% with no quality loss on the tasks that don't need top-tier models.
Hack 7: Install Pearl as Your Smart Router
Pearl (mentioned in Hack 1) does model routing in addition to memory management. Its routing logic is trained on thousands of real OpenClaw usage patterns — it routes better than any manual regex rule you'll write.
clawhub install pearl-router
openclaw config set model_router pearl
Pearl uses cost, latency, and quality signals simultaneously. It knows, for example, that asking for a creative haiku at 2 AM India time is fine on Groq's free tier but that a code review at 9 AM should use Claude because quality matters more than cost.
Hack 8: Set Up Failover Between API Providers
API providers go down. Groq has outages. Anthropic has rate limit spikes. A failover configuration means your agent never silently fails:
# In config.yml
model_fallback_chain:
- claude-3-5-sonnet # Primary
- groq/llama-3.1-70b # First fallback (fast, free)
- ollama/llama3.2:3b # Last resort (local, always available)
With this in place, if Claude's API is unreachable at 3 AM when your overnight cron runs, OpenClaw falls back to Groq, then falls back to your local Ollama model. The task completes. You wake up to results.
Hack 9: Build an India-First Local Model Stack With Ollama
For Indian users who want zero API costs and complete data privacy, the Ollama local stack is the optimal base configuration. Install Ollama, pull the models, and configure OpenClaw to use them:
# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh
# Pull recommended models (current Feb 2026 recommendation)
ollama pull qwen2.5:14b # Best overall quality locally
ollama pull llama3.2:3b # Fast, lightweight for simple tasks
ollama pull nomic-embed-text # For memory embeddings
# Configure OpenClaw
openclaw config set model ollama/qwen2.5:14b
openclaw config set embed_model ollama/nomic-embed-text
openclaw config set api_url http://localhost:11434
Total monthly cost: ₹0. Total data leaving India: ₹0. This is the stack most privacy-conscious Indian users run.
Hack 10: Use Groq's Free Tier as Your Speed Layer
When you need speed and the task is not sensitive, Groq's free tier is genuinely impressive — it can run Llama 3.1 70B at 800+ tokens per second, making it faster than many paid cloud APIs.
openclaw config set groq_api_key YOUR_GROQ_KEY
openclaw config set fast_model groq/llama-3.1-70b-versatile
Route quick lookups, time-sensitive alerts, and high-volume summarisation tasks through Groq to stay within the free tier while reserving Claude for deep reasoning work. As of March 2026, Groq's free tier allows 14,400 requests/day — more than enough for personal use.
Cron & Automation Hacks
OpenClaw's cron system is where the real automation power lives. Most beginners never touch it. Power users make it the backbone of their day.
Hack 11: Cron vs. Heartbeat — the 90% Cost Hack
The single biggest cost mistake beginners make: using a heartbeat (polling every few minutes) instead of a cron (scheduled at a specific time).
A heartbeat that checks your email every 5 minutes makes 288 API calls per day even when nothing has happened. A cron that runs at 8 AM, 1 PM, and 6 PM makes 3 API calls per day. For the same practical result — you see your emails three times a day — crons cost 96 fewer API calls. Scale this across a few automations and you can hit 90% cost reduction.
# heartbeat.yml — avoid this
check:
interval: 5m
task: "Check email and summarise new messages"
# cron.yml — use this instead
morning_email:
schedule: "0 8 * * *" # 8:00 AM IST daily
task: "Summarise all unread emails since yesterday"
afternoon_email:
schedule: "0 13 * * *" # 1:00 PM IST daily
task: "Summarise any urgent emails since this morning"
Hack 12: Build a Dream Routine
A dream routine is a cron that runs overnight while you sleep — preparing everything you need for the next day. Indian power users typically schedule this at 11 PM or midnight IST:
dream_routine:
schedule: "0 23 * * *"
tasks:
- "Check tomorrow's calendar and write a brief for each meeting to meetings/tomorrow.md"
- "Pull today's BSE/NSE closing data for my watchlist stocks and write to market/today.md"
- "Check IRCTC PNR status for any upcoming trips"
- "Summarise unread Slack messages from today"
- "Check weather for tomorrow in Bengaluru and add to morning-brief.md"
When you open your laptop at 7 AM, everything is ready. No morning panic. No context-switching to gather information.
Hack 13: Silent Automations — Tasks That Run Without Pinging You
By default, OpenClaw notifies you when cron tasks complete. For routine automations you trust completely, disable notifications:
silent_file_organiser:
schedule: "0 18 * * 5" # Friday 6 PM
task: "Organise ~/Downloads by file type into dated subfolders"
notify: false # Silent — just does the work
Silent automations are the true measure of trust in your agent. When you start running them without thinking about them, OpenClaw has truly become infrastructure.
Hack 14: Use Shell Filters Before Hitting the LLM
For crons that monitor something (emails, GitHub notifications, news feeds), pre-filter with a shell command before invoking the LLM. This dramatically cuts the number of tokens processed:
github_pr_alerts:
schedule: "0 9,14,18 * * 1-5" # Three times daily, weekdays
task: |
Run: gh pr list --repo myorg/myrepo --state open --json title,author,updatedAt
If there are more than 3 open PRs, summarise them. Otherwise do nothing.
shell_prefilter: "gh pr list --repo myorg/myrepo --state open | wc -l"
skip_if_prefilter_returns: "0\n1\n2\n3"
The shell prefilter runs in milliseconds with zero API cost. The LLM only gets called when there's actually something worth processing.
Hack 15: Trigger Crons via Webhooks for Real-Time Automation
OpenClaw supports webhook triggers — external events can fire your crons immediately rather than waiting for a scheduled time. This is powerful for event-driven workflows:
# Generate a webhook URL for a cron task
openclaw webhook create invoice_processor
# Returns something like:
# Webhook URL: https://your-openclaw.tunnel.dev/webhook/a8f3c2b1
# Now use this URL in your billing software, GitHub Actions, or Zapier
When a new invoice lands in your ~/invoices/pending/ folder (via a Zapier watch), the webhook fires, OpenClaw processes the invoice, fills in GST details, and moves it to ~/invoices/processed/ — all within seconds of the file arriving.
Productivity Hacks
Hack 16: Use Task Queuing for Parallel Work
One of OpenClaw's most underused features is the ability to queue multiple tasks and let the agent work through them while you focus on something else:
openclaw queue add "Write a Python script to parse CSV files from ~/data/"
openclaw queue add "Draft a response to the email from Vikram about the Q1 report"
openclaw queue add "Research the top 5 competitors for our new product feature"
openclaw queue start --parallel 2
The --parallel 2 flag processes two tasks simultaneously. You queue them, walk away, and come back to find three completed tasks. This shifts your relationship with OpenClaw from conversation to delegation.
Hack 17: Run 3 Concurrent OpenClaw Sessions
Many users do not know you can run multiple OpenClaw instances simultaneously, each in a different workspace with a different context:
# Terminal 1 — Personal workspace
openclaw start --workspace ~/openclaw-personal
# Terminal 2 — Work project workspace
openclaw start --workspace ~/openclaw-work-project-atlas
# Terminal 3 — Research workspace
openclaw start --workspace ~/openclaw-research
Keep them separate. Your personal assistant does not need to know about your work code. Your research agent does not need your personal calendar context. Three specialised agents outperform one general agent every time.
Hack 18: Deploy Sub-Agents for Parallel Work
For large, parallelisable tasks, OpenClaw supports spawning temporary sub-agents that each work on a piece of the problem simultaneously and report back:
openclaw task "Analyse the last 6 months of sales data.
Use three sub-agents:
- Agent A handles Jan-Feb
- Agent B handles Mar-Apr
- Agent C handles May-Jun
Merge their findings into a single report."
This can compress a 45-minute sequential analysis into a 15-minute parallel one — a genuine 3x speed improvement on data-heavy tasks.
Hack 19: Build Reusable Skill Templates
Every time you find yourself explaining a recurring task to OpenClaw, that explanation should become a skill. Skills are just Markdown files:
mkdir -p ~/.openclaw/skills
cat > ~/.openclaw/skills/weekly-report.md << 'EOF'
# Skill: Weekly Status Report
## Purpose
Generate a structured weekly status report from my task notes.
## Input
- Read all files in ~/tasks/week-YYYY-WW/
- Read calendar events from this week
## Output Format
Create a file ~/reports/week-YYYY-WW-status.md with:
1. Accomplishments (bullet points)
2. In Progress items
3. Blockers (if any)
4. Next week's priorities
## Notes
- Use past tense for completed items
- Flag anything that slipped from last week's priorities
- Keep it under 400 words — this goes to my manager
EOF
Now whenever you want your weekly report: openclaw skill run weekly-report. No explanation needed.
India Cost Hacks
Hack 20: The Full Local Ollama Stack — ₹0/Month
If you have a reasonably powerful laptop or desktop (16GB+ RAM), you can run a complete OpenClaw setup with zero monthly API costs using Ollama:
| Component | Free Alternative | Monthly Cost | |---|---|---| | LLM (main) | Ollama + Qwen 2.5 14B | ₹0 | | LLM (fast) | Ollama + Llama 3.2 3B | ₹0 | | Embeddings | Ollama + nomic-embed-text | ₹0 | | Hosting | Your own laptop | ₹0 | | Total | Full local stack | ₹0/month |
The trade-off: local models are slightly less capable than Claude Sonnet for complex reasoning tasks. For everyday personal assistant tasks — reminders, file organisation, email drafts, calendar prep — you will not notice the difference.
For comparison, running OpenClaw with Claude Sonnet as your primary model costs approximately ₹1,200-2,500/month in API fees depending on usage volume. The Ollama stack is legitimately free and keeps all your data on Indian soil.
Hack 21: Groq Free Tier Routing for Speed-Critical Tasks
When you need a cloud model but want to stay free, Groq's free tier is the best option available to Indian users right now. As of March 2026:
- Free tier: 14,400 requests/day, 6,000 requests/minute
- Models available: Llama 3.1 8B, Llama 3.1 70B, Mixtral 8x7B, Gemma 2 9B
- Speed: 800+ tokens per second — fastest inference available
- India availability: Full access, no restrictions
- Payment required: No (just a Groq account)
The combined strategy used by the most cost-conscious Indian OpenClaw users:
- Local Ollama stack for private/offline tasks
- Groq free tier for fast cloud tasks
- Claude API reserved only for tasks that genuinely need frontier model intelligence
Result: 95%+ of tasks handled at ₹0/month, with Claude API costs under ₹200/month for the rare cases that need it.
Get Started With OpenClaw
These 21 hacks compound. Start with Hack 2 (ABOUT.md) and Hack 11 (crons over heartbeats) — both take under 10 minutes and immediately improve your experience. Then layer in Pearl (Hacks 1 and 7) for automatic memory and model routing.
If you are just getting started with OpenClaw, read our complete setup guide first: OpenClaw Setup Guide for India.
Ready to explore the full skill ecosystem? Browse 13,700+ skills on ClawHub to find tools for your specific workflow.
The best OpenClaw setup is the one that reflects how you actually work — not how someone else works. Start with these hacks, adapt them to your context, and your agent will become genuinely irreplaceable.