💡 Use Cases
Real-world examples and success stories
36-Hour Ralph Loop: Opus Orchestrates Qwen 3.6/3.8 Builders on M1 Max
A developer is running a 36-hour /goal Ralph loop: Opus 5.5 orchestrates Qwen 3.8 27B 4-bit and Qwen 3.6 35B A3B 8-bit builders on two M1 Max 64GB machines to ship 38 features and 201 tasks.
Apple's On-Device 3B Model as a Checker for OpenClaw Agent Output
A developer wires Apple's on-device 3-billion-parameter model into OpenClaw as a fact-checker that validates agent answers against tool results before output.
OpenClaw Automates DLSS Modding for Cyberpunk: Day of DLL Hunting to 30 Minutes
A developer used OpenClaw to handle DLSS mod installation on an RTX 4070. What took a full day of DLL hunting and manual setup now takes about 30 minutes, with GPT-6.1-Sol on max thinking doing the research.
Dev Replaces $239.99/Year Plaud Subscription With Local Transcription Skill
A developer replicated the Plaud device's paid transcription service using a locally-run app, skill, and cron job — skipping the $239.99/year subscription. Transcripts land on the local machine where they can be queried across past meetings.
War Radar: An Automated Pipeline That Uses AI to Classify Conflict and Evacuation Reports
War Radar publishes a daily world map of armed-conflict risk: a formula sets the risk level, AI models handle only narrow classification, and the project publishes how often each level was followed by a crisis.
OpenClaw Yolo Mode: Autonomous Agent Loop on a €20/Month Budget
A user wired OpenClaw into a Yolo Mode loop where a main agent assigns work to a Project Manager sub-agent, checks subscription usage hourly, and pauses everything below 20% until the limit resets.
OpenClaw Agent Broke at 65% Context: 683k Tokens, Zero Cache Reads on Ollama/GLM
A 23-hour OpenClaw Discord agent run on GLM 5.3 Flash via Ollama Cloud broke at 683k tokens — 65% of the configured 1M window — with no overflow error, no timeout, and cacheRead=0 on every affected turn.
Novice Builds Local 'Second Brain' on Intel Arc Pro B60: Qwen 3.8 27B at 38 tok/s, 128K Context
A non-developer runs OpenClaw as a personal agent on an old Dell, then builds a separate AI server around an Intel Arc Pro B60 serving Qwen 3.8 27B at ~38 tok/s with a 128K context window.
QWEN3.6-27B-MLX-8bit Matches 122B for Local Workflows at 29.5GB
A user running multi-hour local workflows says QWEN3.6-27B-MLX-8bit (29.5GB) matched a 79GB model on orchestration, research, and writing — just slower — freeing 50GB of RAM.
OpenClaw Agent's Plain MEMORY.md Setup Beats Memory Startup's Runtime in Temporal Test
A memory startup's runtime returned the wrong version in a temporal-decision test, tying all three at 1.000 relevance. The user's OpenClaw agent, with memory as markdown in a git repo, returned the current decision dated.
Hermes vs OpenClaw: Benchmark on Real Work Ends with a Memory Lesson
A user benchmarked Hermes and OpenClaw on real tasks, finding performance similar but OpenClaw's explicit memory won them over after Hermes over-generalized a passing compliment.
Triaging Community PRs with Multiplayer Agents in OpenClaw 2.0
Patrick Erichsen (OpenClaw Foundation) shows how to use multiplayer agents in OpenClaw 2.0 to triage and review community PRs. The automation checks Discord every 15 minutes, assigns a maintainer, and starts a shared agent session.