OpenClaw Performance Reports

Public performance report archive for OpenClaw.

This repository is written by the OpenClaw Performance GitHub Actions workflow. It stores generated benchmark reports, source-probe summaries, and latest-run pointers so performance changes on main can be inspected without downloading workflow artifacts.

Latest Reports

• Mock provider: deterministic fake OpenAI-compatible provider, focused on OpenClaw loop overhead.
• Mock deep profile: CPU, heap, and trace artifacts for startup, Gateway, and agent-turn hotspots.
• Live GPT-5.4: real openai/gpt-5.4 agent turn when the OpenAI key is available to CI.

Each latest pointer contains the source repository, commit SHA, workflow run id, lane, and report path.

Layout

Reports are grouped by source ref, workflow run, attempt, and lane:

openclaw-performance/<ref>/<run-id>-<attempt>/<lane>/

Typical files:

• index.md: human-readable lane summary.
• report.md: Kova report summary.
• report.json: raw Kova report data.
• source/index.md: OpenClaw source-probe summary when available.
• source/*.json: raw source-probe data when available.

What Is Measured

Current reports include:

• Gateway boot timing and memory.
• Startup with default, hook, and plugin-heavy configurations.
• Fake-provider agent turns for pure loop overhead.
• Real GPT-5.4 agent turn latency when live credentials are present.
• CLI startup commands against a booted Gateway.
• CPU, heap, and trace artifacts for hotspot investigation.

The benchmark harness lives in openclaw/Kova. The workflow configuration lives in openclaw/openclaw.