Skip to main content
Most support bots are FAQ search bars with a chat interface. User asks a question, bot pattern-matches against a static list, returns a canned answer. User comes back tomorrow and the bot has no idea who they are. That’s not support. That’s a worse version of Ctrl+F. An OpenClaw support system is a set of persistent agents that remember every user, maintain their own knowledge bases on disk, use real tools to investigate account issues, and autonomously escalate when they hit their limits. And they get smarter over time because they write what they learn back to their filesystem.

​
Why this actually works

It remembers every user. When someone comes back with a follow-up three days later, the agent knows their previous issue, what was tried, and what the resolution was. No “can you describe your issue again?” Ever. It maintains a knowledge base on disk. FAQs, troubleshooting guides, product docs, known issues. All of it lives on the instance’s filesystem. When your product changes, update the files and the agent reads them on every interaction. Better yet, have the agent write to disk too: common issues it encounters get logged to knowledge/common-issues.md so patterns surface over time. The knowledge base grows itself. It uses tools to investigate. Account lookup skills to check a user’s subscription status, order history, or configuration. Web search to find relevant docs or status pages. The agent does real investigation instead of just pattern-matching against a FAQ. It escalates on its own. When the agent determines it can’t resolve an issue, it escalates autonomously. It formats the conversation history, diagnostic results, and attempted solutions into a structured ticket for the next tier. The user doesn’t have to do anything. They don’t even know the handoff happened until a more specialized agent (or a human) picks up.

​
The setup

Create three instances for three support tiers: Tier 1: First response. Connected to your public Discord server or Slack workspace. Handles common questions by reading the FAQ and product docs from its filesystem (knowledge/faq.md, knowledge/product-guide.md). Uses account lookup skills to check user status before responding. When it resolves an issue, it logs the solution to knowledge/resolutions.md. When it can’t help, it autonomously escalates by sending the conversation summary and diagnostic data to the Tier 2 channel. Tier 2: Technical support. Connected to a separate Discord channel or Telegram bot. Has a deeper filesystem: docs/api-reference.md, docs/integration-guide.md, docs/known-issues.md, diagnostic flowcharts. Uses code execution skills to reproduce bugs, web search to check for related issues in public forums, and file handling to read and update the known issues list. When it resolves something new, it writes the solution back to its knowledge base so it handles it autonomously next time. Tier 3: Human handoff. A lightweight instance on Slack that your support team monitors. Receives autonomously escalated tickets from Tier 2, formatted with full context: user info, issue summary, conversation history, diagnostic results, and what was already tried. Your human agents pick up without asking the user to repeat anything.

​
How this compounds

Week 1. User reports a login bug. Tier 1 checks their account status via lookup tool, walks through standard troubleshooting from knowledge/faq.md, resolves it. Logs the resolution. Week 2. Same user comes back with a different issue. Tier 1 remembers them, knows their account details, skips the questions it already asked. Checks their configuration using account tools. This time it’s a deeper issue. The agent autonomously escalates to Tier 2 with full context. Week 3. A third user reports the same login bug from Week 1. Tier 1 reads its resolution log, recognizes the pattern, resolves it in one message. What took 5 minutes in Week 1 takes 30 seconds in Week 3. Month 2. Tier 2 has resolved 50 technical issues. Its known-issues.md file is a living document the agent maintains itself. Many issues that would have escalated to Tier 2 in Month 1 are now handled by Tier 1 because solutions were written back to the knowledge base. Your support system is literally teaching itself.

​
What to configure

​
Filesystem per tier

  • Tier 1. knowledge/faq.md, knowledge/product-guide.md, knowledge/resolutions.md (agent-maintained), knowledge/common-issues.md
  • Tier 2. docs/api-reference.md, docs/integration-guide.md, docs/known-issues.md (agent-maintained), diagnostic flowcharts, architecture docs
  • Tier 3. Minimal. templates/ticket-format.md for structuring escalated tickets

​
Skills

  • Account lookup for Tier 1 and 2 to check subscription status, order history, configuration
  • Web search for Tier 2 to check public forums, status pages, and documentation
  • Code execution for Tier 2 to reproduce bugs and test fixes
  • File handling for all tiers to read their knowledge base. Tier 1 and 2 write back resolutions and patterns.

​
Personas

  • Tier 1. Friendly, concise. Reads FAQ before every response. Knows when to stop guessing and escalate. Logs resolutions to disk.
  • Tier 2. Technical, methodical. Follows diagnostic workflows. Updates the known issues file when it resolves something new.
  • Tier 3. Minimal. Formats incoming escalations into clean tickets for humans.

​
Escalation behavior

Define in each persona’s system prompt when and how to escalate. Tier 1 escalates when it can’t answer from its knowledge base after checking account data. Tier 2 escalates when an issue requires human judgment (refunds, policy exceptions, edge cases). Escalation is autonomous. The agent sends the formatted summary to the next tier’s channel without requiring the user to do anything. Three tiers fit on the Pro plan. Per-product or per-region support tiers on the Max plan.