Clawdbot's Memory System: Built on Markdown

OpenClaw (formerly Clawdbot) uses a unique local-first memory system built entirely on Markdown files.

Clawdbot (now renamed OpenClaw) — The experience is truly impressive, especially its persistent memory management system — it maintains context 24/7, remembers conversations indefinitely, and continuously deepens interactions based on that.

Regarding Clawdbot’s memory management system, AI research engineer Manthan Gupta wrote a detailed article dissecting the unique principles behind Clawdbot’s memory mechanism.

“It doesn’t use the cloud-based, big-company-controlled memory approach; instead, it keeps everything local, giving users complete control over their own context and skills.”

All memories are archived and managed using Markdown (.md) documents.

How are context and memory defined in Clawdbot?

1.1 How is context constructed?

Before diving into memory, we first need to understand what the model actually “sees” when processing each request:

[0] System Prompt (static + conditional instructions) 
[1] Project context (guidance files: AGENTS.md, SOUL.md, etc.) 
[2] Conversation history (messages, tool calls, compressed summaries) 
[3] Current message