data-worker<\/code> agents work together.<\/p>\n
\nCommon Token Optimization Mistakes<\/h2>\nMistake 1: Using Opus for Batch Jobs<\/h3>\n
Opus ($15\/1M tokens) is a reasoning powerhouse. But if you’re processing 50 URLs, extracting titles, or running classification tasks \u2014 that’s Flash territory. We’ve seen batch jobs that should cost $0.15 cost $22 because the model wasn’t switched.<\/p>\n
Fix<\/strong>: Pin batch and automation tasks to fast<\/code>. Only escalate if the output quality is measurably bad.<\/p>\nMistake 2: Timestamps in System Prompts<\/h3>\n
If your workspace files inject “Current Date and Time: March 2, 2026 14:30:05” into the system prompt, you’ve just invalidated your entire cache. Every call gets a unique prefix. Every call pays full price.<\/p>\n
Fix<\/strong>: Keep workspace files 100% static. Let the model infer the date from conversation context, or inject it in the user message (not the system prompt).<\/p>\nMistake 3: Never Compacting Sessions<\/h3>\n
OpenClaw sessions grow. A 50-turn conversation about SEO analysis can hit 100K+ tokens. Every subsequent message pays for all that context.<\/p>\n
Fix<\/strong>: Run \/compact<\/code> after completing each task. Start \/new<\/code> sessions between unrelated tasks. Check \/status<\/code> regularly \u2014 if context exceeds 30K for a simple task, compact immediately.<\/p>\nMistake 4: Loading Entire Files as Memory<\/h3>\n
Without QMD, OpenClaw dumps your entire MEMORY.md, all daily notes, and any referenced files directly into context. A 5,000-token memory file is loaded in full even when the conversation only needs one paragraph.<\/p>\n
Fix<\/strong>: Install QMD. It returns only the 5 most relevant snippets instead of the entire file. 90% reduction in memory tokens.<\/p>\nMistake 5: Not Setting a Budget Guardrail<\/h3>\n
“I’ll monitor it manually” works until it doesn’t. One unattended batch job at 3 AM can blow your monthly budget in a single night.<\/p>\n
Fix<\/strong>: Set a $3\/day guardrail on OpenRouter immediately. Takes 30 seconds. Prevents the $200 surprise bills that show up on forums regularly.<\/p>\n
\nThe Complete Cost Breakdown<\/h2>\n
Here’s what our operation actually costs with all optimizations applied:<\/p>\n
\n
\n\n\n| Category<\/th>\n | % of Tasks<\/th>\n | Model<\/th>\n | Monthly Cost<\/th>\n<\/tr>\n<\/thead>\n |
\n\n| Heartbeats & idle checks<\/td>\n | ~15%<\/td>\n | Gemini Flash<\/td>\n | ~$1.50<\/td>\n<\/tr>\n |
\n| Data extraction & file ops<\/td>\n | ~40%<\/td>\n | Gemini Flash<\/td>\n | ~$4.00<\/td>\n<\/tr>\n |
\n| SEO analysis & browsing<\/td>\n | ~15%<\/td>\n | Kimi K2.5<\/td>\n | ~$5.00<\/td>\n<\/tr>\n |
\n| Content writing<\/td>\n | ~25%<\/td>\n | Claude Sonnet<\/td>\n | ~$14.00<\/td>\n<\/tr>\n |
\n| Architecture & debugging<\/td>\n | ~5%<\/td>\n | Claude Opus<\/td>\n | ~$3.00<\/td>\n<\/tr>\n |
\nTotal<\/strong><\/td>\n100%<\/strong><\/td>\nMixed<\/strong><\/td>\n~$27.50<\/strong><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n\n MONTHLY COST COMPARISON<\/p>\n $27 vs $87<\/p>\n Same quality output. Same number of tasks. Just smarter routing.<\/p>\n<\/div>\n The writing budget ($14\/month on Sonnet) is non-negotiable. Content quality is what drives SEO rankings. You save everywhere else so you can afford to spend here.<\/p>\n \n“The teams that understand model routing will build 10x more with the same budget. It’s not about spending less \u2014 it’s about spending on the right tokens.”<\/p>\n \u2014 Matt Ganzak, OpenClaw Token Optimization Guide, 2026<\/p>\n<\/blockquote>\n
\nSecurity Guardrails (Non-Negotiable)<\/h2>\nToken optimization shouldn’t compromise security. These are the guardrails we never disable:<\/p>\n \n- 🔒 Heartbeat uses API model with prompt injection resistance (not local Ollama)<\/li>\n
- 🔒 Gateway bound to
127.0.0.1<\/code> only \u2014 never exposed to the network<\/li>\n- 🔒 Token-based gateway authentication<\/li>\n
- 🔒 Phone\/user allowlist on messaging channel<\/li>\n
- 🔒 HEARTBEAT.md kept empty \u2014 minimal attack surface during heartbeat cycles<\/li>\n
- 🔒 Never use local models for tasks involving untrusted content (web scraping, email processing)<\/li>\n<\/ul>\n
Related<\/strong>: Securing Your AI Agent in 2026: ClawHavoc & CVE-2026-25253<\/i><\/a> covers the ClawHavoc supply chain attack (1,184 malicious skills), the WebSocket RCE, and how we hardened against them.<\/p>\n
\nFrequently Asked Questions<\/h2>\nIs OpenClaw free to run?<\/h3>\nOpenClaw itself is free and open-source.<\/strong> The cost comes from the AI models it calls through APIs like OpenRouter. With our optimized config, expect $18-27\/month for a production SEO operation. A minimal personal assistant setup can run under $5\/month on Gemini Flash alone.<\/p>\nCan I use Ollama to make it completely free?<\/h3>\nTechnically yes, but we don’t recommend it for production.<\/strong> Local models lack the prompt injection hardening of API models. For a personal hobby project with no sensitive data, Ollama is fine. For a business operation handling credentials, emails, and financial data \u2014 use API models with training-level security. Gemini Flash at $0.10\/1M tokens is nearly free anyway.<\/p>\nHow much does Claude Opus cost on OpenClaw?<\/h3>\nAbout $15 per million tokens.<\/strong> In our setup, Opus handles roughly 5% of tasks (architecture decisions, complex debugging, security audits), costing about $3\/month. The key is never letting Opus touch routine tasks. A single batch job accidentally routed to Opus can cost more than your entire month of Flash usage.<\/p>\nDoes prompt caching work with OpenRouter?<\/h3>\nYes, with caveats.<\/strong> Set cacheRetention: \"long\"<\/code> for a 1-hour TTL. Cached reads get a 90% discount on Anthropic models. However, some OpenRouter provider pass-throughs don’t forward cache_control<\/code> headers properly (see GitHub Issue #9600). Verify by checking that cacheRead > 0<\/code> after multiple turns in the same session.<\/p>\nWhat’s the minimum setup for token optimization?<\/h3>\nThree changes that take five minutes:<\/strong><\/p>\n\n- Set
\"model\": \"google\/gemini-2.0-flash-001\"<\/code> as default in openclaw.json<\/li>\n- Set
\"heartbeat.model\": \"google\/gemini-2.0-flash-001\"<\/code><\/li>\n- Set a $3\/day budget guardrail on OpenRouter<\/li>\n<\/ol>\n
That alone cuts 50-60% off most users’ bills. Add caching, QMD, and per-agent configs later for the remaining savings.<\/p>\n How do I check if my optimizations are working?<\/h3>\nRun \/status<\/code> in any OpenClaw conversation.<\/strong> It shows your current model, context size, and token usage. Then check the OpenRouter dashboard for per-model spending breakdown. After 24 hours, verify: heartbeats hit Flash (not Sonnet\/Opus), writing tasks hit Sonnet, and daily spend stays under $3.<\/p>\n
\nGetting Started: Your Next Steps<\/h2>\n\n ☑ Quick-Start Checklist<\/h3>\n | | | |