OpenClaw OptimizationPatterns for High-Performance Agents

Practical strategies from real autonomous operation — token efficiency, behavioral compliance, cron reliability, and multi-agent coordination. A living document that evolves with production experience.

Getting Started

The fundamentals — workspace rules, model routing, memory search, cron basics, and realistic expectations. Start here.

Start here →

Workspace Files & Bootstrap

What belongs in each file, instruction format, the ownership problem, token cost at scale, truncation risks, and light context mode.

Workspace files →

Memory Search

Local embeddings, hybrid vector + FTS queries, temporal decay, and memory maintenance. The highest-impact infrastructure to set up early.

Memory search →

Model Optimization

Match model capability to task complexity. Thinking parameter tuning, fallback chains, and the real cost math behind model selection.

Model optimization →

Scheduling & Automation

Gateway crons, sentinel watchers, and native schedulers — choosing the right tool, cost optimization, circuit breakers for reliability.

Scheduling →

Agent Architecture

Multi-agent identity sharing, directory-based coordination, tool lockdown, workspace isolation, and context injection.

Architecture →

Agentic Patterns

Scout/dispatch, specialized subagents, research fleets, multi-stage pipelines, fire-and-forget, and orchestration transparency.

Agentic patterns →

Self-Improvement

How an agent learns from mistakes without script churn. Daily notes, lesson promotion, automated extraction, and human-in-the-loop review.

Learning loops →

Behavioral Quality

Response calibration, streaming responsiveness, channel-appropriate formatting, and mid-run interruption.

Behavior →