swarm

What This Skill Does

Swarm is a high-performance orchestration layer for OpenClaw that enables massive cost reduction by offloading parallel, batch, and research tasks to Gemini Flash workers. Rather than utilizing your primary LLM for every sequential step, Swarm distributes tasks across a distributed swarm of workers, slashing costs by up to 200x while simultaneously reducing execution time from minutes to seconds. It supports sophisticated execution modes including Parallel, Research, Chain, and Benchmarking, effectively turning your expensive primary model into an affordable daily driver.

Installation

To integrate Swarm into your environment, use the OpenClaw hub CLI tool: clawhub install openclaw/skills/skills/chair4ce/swarm

Ensure that you have your local daemon configured correctly by running swarm status after installation to verify the connection to the worker network.

Use Cases

• Massive Research: Use the Research mode to perform multi-phase search-fetch-analyze operations across multiple subjects simultaneously.
• Batch Processing: Process thousands of entities, facts, or document fragments in parallel to maximize throughput.
• Refinement Pipelines: Implement complex data flows using the Chain mode, passing data through specific filters like 'analyst', 'critic', or 'strategist' for multi-stage refinement.
• Cost-Optimized Benchmarking: Run 'LLM-as-a-judge' experiments to compare output quality across different configurations without incurring high API costs.

Example Prompts

1. "Swarm, research the pricing structures of the top 5 CRM platforms for 2026 and summarize the key differences in a table."
2. "Use swarm parallel mode to explain the concept of quantum entanglement to a child, a teenager, and a PhD student simultaneously."
3. "Start a chain with a 'researcher' and an 'analyst' stage to investigate the current market trends in renewable energy and generate a synthesis report."

Tips & Limitations

• Check Daemon: Always run swarm status at the start of your session to ensure the orchestrator is reachable.
• Depth Selection: When using Chain mode, start with the 'quick' preset to avoid over-engineering; only scale to 'exhaustive' (8 stages) for critical, high-stakes tasks.
• Avoid Overloading: While highly efficient, Swarm is best suited for discrete, independent tasks. If a task requires absolute, strict sequential memory of previous outputs, ensure your Chain pipeline is configured correctly for state passing.
• Safety: Since Swarm triggers external API calls for research and grounding, ensure your environment handles network-accessible tools according to your security policy.