Natural language rules

DlxAI's killer feature: no code, no flowcharts. Describe what you want in a single sentence (English or Chinese). DlxAI compiles it into the right policy, guard, or skill, and applies it instantly.

Three kinds of artefact

• Policy: Always-on behavioural constraint. e.g. "never proactively contact strangers"
• Guard: Pre/post checks around tool calls. e.g. "confirm before running shell commands"
• Skill: A triggerable capability. e.g. "summarise yesterday's messages every morning at 9"

The compilation is transparent — you only write plain language.

Examples

Scheduled task

Every morning at 9, summarise yesterday's Telegram messages and send to me.

Compiles to: cron trigger (0 9 * * *) + fetch_messages skill + summarisation LLM call.

Triggered response

When a customer messages "quote" in Lark, look up the latest pricing sheet and reply.

Compiles to: Lark keyword trigger + knowledge-base lookup skill + auto-reply.

Guard

Never let the agent read any file under ~/.ssh.

Compiles to: file_permissions guard injected into OpenClaw's before_tool_call hook.

Policy

Keep iMessage replies short — at most two sentences.

Compiles to: channel-scoped system prompt injection.

How to write good rules

• Be specific: "scheduled" → "every day at 9am"; "auto reply" → "reply when keyword X appears"
• Spell out the trigger: "When X happens, do Y"
• Scope it: "in Telegram", "to customer groups", "not in DMs"
• Give an example if useful: "e.g. on receiving 'hi', reply 'welcome'"

Rule not firing?

If a rule isn't behaving as expected, check the gateway log panel on the left. The compiled artefact for each rule is also visible in the rule's detail page for debugging.