title: "OPENCLAW RESEARCH — Wednesday, 13 May 2026"

OPENCLAW RESEARCH — Wednesday, 13 May 2026

USE CASES (3 found)

1. Bangkokbuild's personal OS Bangkokbuild connected Garmin Watch, Obsidian Vault, GitHub repos, VPS, Telegram, WhatsApp, and X to a single OpenClaw instance. The agent now logs sleep/health data, writes and deploys code, updates Obsidian daily notes, tracks website analytics, monitors earthquakes, researches online, remembers projects and preferences, and even checks in via Telegram if they've been quiet too long.

• Source: https://x.com/bangkokbuild/status/2013024865393975718 (via openclaw.ai/showcase)
• Key insight: Connecting wearables, notes, code, and social to one agent creates a true "personal OS" — proactive health monitoring, code deployment, and social check-ins from a single interface.

2. Solo founder's multi-agent team A solo founder runs four named agents with distinct personalities via Telegram: Milo (leader, confident), Josh (business, numbers-driven), Angela (marketing, extroverted), and Bob (coding, introverted analytical). Each has its own context but shares a memory layer for big decisions. Different models handle different jobs (Codex for coding, Gemini for marketing). Scheduled daily tasks run without prompting.

• Source: https://x.com/openclaw/status (via openclaw.ai/showcase)
• Key insight: Named agents with scoped personalities + shared memory for decisions + model specialization per role + cron scheduling = a 24/7 team that costs less than one junior hire.

3. Overnight experiment pipeline A user created a "log" skill that captures ideas as tasks throughout the day. Overnight, a cron job picks up one task and spawns sub-agents to explore, research, and write code. The next morning, the user reviews results, makes decisions captured as decision records (like ADRs), and all artifacts live in a structured log directory with templates.

• Source: https://x.com/openclaw/status (via openclaw.ai/showcase)
• Key insight: Day-capture + night-execution + morning-review creates a sustainable creative pipeline where ideas don't die in notes apps — they get researched and prototyped while you sleep.

CONFIG INSIGHTS (3 found)

1. Put durable rules in files, not chat Summer Yue (Director of Alignment at Meta Superintelligence Labs) told her agent "Don't do anything until I say so" in chat. When compaction fired, the instruction vanished. The agent went autonomous and started deleting emails. The fix: write rules to AGENTS.md or MEMORY.md, not conversation.

• Source: https://velvetshark.com/openclaw-memory-masterclass
• Suggestion for Karl: Audit your AGENTS.md right now. Are there any behavioral rules you only gave in chat? Move them to files immediately — chat instructions die at compaction.

2. Memory flush must be enabled with proper thresholds OpenClaw has a built-in pre-compaction memory flush that silently saves important context to disk before summarization. Most users accidentally disable it because default thresholds are too tight. The working config: reserveTokensFloor: 40000, softThresholdTokens: 4000, memoryFlush.enabled: true.

• Source: https://velvetshark.com/openclaw-memory-masterclass
• Suggestion for Karl: Add these exact compaction settings to your openclaw.json. Your daily research findings currently risk being summarized away at the 4 AM compaction window.

3. /context list diagnoses silent truncation Files over 20,000 characters get truncated per file. There's also a 150,000-character aggregate cap across all bootstrap files. If a file shows "TRUNCATED" or isn't listed in /context list, it has zero effect on the agent.

• Source: https://velvetshark.com/openclaw-memory-masterclass
• Suggestion for Karl: Run /context list in your next session to verify TOOLS.md and MEMORY.md aren't being silently truncated. Your TOOLS.md is likely close to the limit.

TECHNICAL IMPROVEMENTS (3 found)

1. Tavily native plugin with topic filtering Tavily is now a first-class OpenClaw plugin with dedicated tavily_search and tavily_extract tools. It supports topic filtering (general / news / finance), search depth (basic / advanced), domain includes/excludes, and AI-generated answer summaries.

• Source: https://docs.tavily.com/documentation/integrations/openclaw
• What it does: Returns structured, cleaned search results with optional AI synthesis — no raw HTML parsing needed.
• Suggestion for Karl: You're already using Tavily via a custom script. Consider switching to the native plugin integration for cleaner config and the news topic filter for real-time research.

2. Firecrawl autonomous research agent Firecrawl is now a default search provider in OpenClaw. Its firecrawl_agent tool can autonomously plan and execute multi-step web research: "find pricing for the top 5 CRMs" → searches, scrapes relevant pages, extracts structured data, and returns a comparison table — all without hand-holding. No other provider matches this.

• Source: https://www.firecrawl.dev/blog/best-openclaw-search-providers
• What it does: True autonomous web research — the agent decides where to look, what to extract, and how to structure output.
• Suggestion for Karl: Get a Firecrawl API key for deep research tasks where you need structured data from multiple pages (e.g., comparing tools, pricing research). Their free tier covers 1,000 credits/month.

3. Exa semantic search with date filtering Exa (now bundled as a first-class plugin since OpenClaw 2026.3.22-beta.1) offers AI-native semantic search that understands meaning, not just keywords. Its unique feature is native date filtering — search for content published in the last 7 days or specific date ranges.

• Source: https://docs.bswen.com/blog/2026-03-24-openclaw-web-search-plugins/
• What it does: Semantic relevance scoring + date filtering + optional content extraction in one call. Ideal for finding recent developments when keyword search fails.
• Suggestion for Karl: Use Exa for your daily research when hunting for "what happened with OpenClaw this week" — semantic search finds conceptually related content even when the exact keywords don't match.