openclaw-config

Battle-tested configs, scripts, and workspace templates for OpenClaw.

• Upstream: https://github.com/openclaw/openclaw
• This repo is not an SDK or plugin - it's a set of copyable building blocks.

What you get

• Memory Engine (scripts/memory-engine/): capture → recall → decay → learn (nightly) + self-review (MISS/FIX) from real logs
• Gateway config snippets (config/*.json5): compaction, pruning, model fallbacks, Discord setup, memory search
• Workspace templates (templates/): AGENTS.md, SOUL.md, HEARTBEAT.md, etc.
• Optional content pipeline (content-pipeline/): cron-driven draft → approval → post → metrics loop

What's New (Feb 2026)

🚀 New Config Examples:

• pro-optimization.json5 - Production-tuned power user settings (the "dial it to 11" config)
• feature-unlocks.json5 - Enable the "60% to 100%" features that ship disabled by default
• memory-lancedb-setup.json5 - LanceDB-backed long-term memory with auto-capture/recall
• web-search-perplexity.json5 - Switch from Brave to Perplexity (no rate limits)

📚 New Docs:

• diagnosis-protocol.md - Check tools before recommending workarounds
• context-overflow-prevention.md - Rules for spawning sub-agents under heavy load

🛠️ New Scripts:

• post-update-macos-arm64.sh - Reinstall native dependencies after npm update

📊 Updated with Real Production Values:

• Compaction & Pruning - Updated with actual running config: reserveTokensFloor: 20000, softThresholdTokens: 8000
• Context Overflow Prevention - Balanced production settings (not aggressive test config)
• Model Setup - Current production model stack: Opus primary, Sonnet + GPT-5.2 fallbacks, Kimi for subagents
• Memory Recall - Increased to 6 results per context (0.4% of window, better continuity)
• Cron Concurrency - Increased to 5 (morning window has overlapping jobs)
• Subagent Thinking - Changed to "high" (subagents do real work)

Why These Matter: OpenClaw ships conservative by default - it prioritizes "just works" over "maximum power." The new configs represent stable, production-tested optimizations that power users want but newcomers don't need to think about. Not experimental. Not bleeding edge. Just the settings that work when you know what you're doing.

Who this is for

You'll like this repo if you:

• run OpenClaw daily and want your setup to be more stable under heavy tool usage
• are tired of "self-improvement" prompts that don't stick
• want a practical starting point for a real workspace (templates + scripts), not a blank folder

Safety / privacy (read before copying)

These files are meant to be copied into your workspace, which means you should treat them like production config:

• Never commit secrets (API keys, tokens, cookie DBs)
• Skim any file before you copy it; some templates contain placeholder personal details
• If you fork this repo: keep it clean (no private data, no logs)

Quickstart (recommended)

Pick only what you want. Most people start with templates + memory engine.

# 1) Clone
git clone https://github.com/unisone/openclaw-config.git
cd openclaw-config

# 2) Copy templates into your workspace (edit them after)
# Replace ~/clawd with your OpenClaw workspace path.
rsync -av templates/ ~/clawd/

# 3) Copy the memory engine
mkdir -p ~/clawd/scripts/
rsync -av scripts/memory-engine/ ~/clawd/scripts/memory-engine/
chmod +x ~/clawd/scripts/memory-engine/*.sh

# 4) Run an initial capture (creates/updates memory/store.json)
cd ~/clawd
bash scripts/memory-engine/capture.sh

# 5) Try recall
bash scripts/memory-engine/recall.sh "context overflow"

Add the nightly job

Run nightly learning (capture → decay → self-review → insights):

• Option A: add a cron job in OpenClaw that runs bash scripts/memory-engine/learn.sh
• Option B: run it manually at first to confirm it matches your workflow

Gateway config snippets

Config examples live in config/*.json5.

Typical workflow:

1. Open your OpenClaw config file (~/.openclaw/openclaw.json on most installs)
2. Copy the snippet you need
3. Merge it into your config (don't blindly replace)

Start here:

• config/compaction-and-pruning.json5 — highest impact stability settings
• config/context-overflow-prevention.json5 — defensive settings for heavy tool use
• config/model-aliases-and-fallbacks.json5 — provider fallbacks
• config/memory-search.json5 — semantic search over memory files
• config/memory-lancedb-setup.json5 — LanceDB + OpenAI embeddings + auto-capture/recall
• config/web-search-perplexity.json5 — Perplexity search (no rate limits with paid key)
• config/discord-setup.json5 — practical Discord defaults

Power User Configs (NEW)

pro-optimization.json5 — The "dial it to 11" config. Every optimization that makes OpenClaw faster, smarter, and more capable without sacrificing stability. What changes:

• Longer session idle time (12h vs 24h for Discord)
• Faster auth recovery (2h billing backoff vs 24h)
• Aggressive caching (15min search, 10min web fetch)
• Subagent cleanup (60min archive)
• Smart message handling (auto-remove 👀 reactions)
• Low thinking by default

When to use: You understand the config, you're running OpenClaw daily, you want maximum capability.

feature-unlocks.json5 — The "60% to 100%" config. Enables production-ready features that ship disabled to avoid overwhelming new users. What unlocks:

• Session transcript search (memory search covers ALL past conversations)
• Image/audio auto-analysis (no need to ask "what's in this image?")
• Link auto-fetch (paste a link, agent reads it — up to 3 links, 15s timeout)
• Typing indicator (shows when agent is thinking)
• Heartbeat active hours (proactive checks during work hours)
• Hot reload (auto-reload config changes without restart)

When to use: You've read AGENTS.md, you have HEARTBEAT.md configured, you want the full power-user experience.

How to apply:

# Read the configs, copy sections you want into ~/.openclaw/openclaw.json
# Restart gateway after changes
openclaw gateway restart

Pro tip: Apply pro-optimization.json5 first, run for a few days, then add feature-unlocks.json5. Changing everything at once makes it hard to tell what broke if something goes wrong.

Repo layout

openclaw-config/
  agents/               # Multi-agent persona prompts
  config/               # JSON5 snippets you can merge into your gateway config
  content-pipeline/     # Optional: cron-driven content workflow
  docs/                 # Deep dives (e.g., context overflow prevention)
  scripts/memory-engine/# Capture/recall/decay/learn + self-review
  templates/            # Workspace bootstrap files
  workflows/            # Optional Lobster workflows

Contributing

PRs welcome, but keep it strict:

• Must be tested in a real OpenClaw workspace
• No personal data (keys, usernames, emails, logs)
• Explain why a setting/script exists, not just what it does

See CONTRIBUTING.md.

License

MIT (see LICENSE)