Put instructions in OpenClaw's workspace
OpenClaw stores memory as Markdown files in its agent workspace, which defaults to ~/.openclaw/workspace. It remembers content saved to disk, with no hidden memory state. OpenClaw memory documentation
Write stable preferences in USER.md
Use USER.md for stable preferences, communication style, relationships, and active-project context. Write profile facts as imperative directives, with observed-date and active or superseded metadata. OpenClaw memory documentation
When a preference changes, supersede its earlier active directive instead of keeping contradictory active instructions. OpenClaw memory documentation
Keep durable decisions in MEMORY.md
MEMORY.md holds durable non-profile facts and decisions, and loads at session start. Keep it curated with short summaries rather than using it as a transcript or exhaustive archive. OpenClaw memory documentation
Use daily notes for changing context
Store running context and observations in dated files such as memory/YYYY-MM-DD.md. Today's and yesterday's notes load automatically after a bare /new or /reset. Daily notes are indexed for memory_search and memory_get, but are not injected into the bootstrap prompt on every turn. OpenClaw memory documentation
The default dreaming sweep can distill useful daily-note material into MEMORY.md. If that file exceeds its bootstrap budget, OpenClaw keeps it on disk but truncates the copy injected into context. OpenClaw memory documentation
Ask OpenClaw to remember something
You can ask OpenClaw to remember information, and it writes the note to an appropriate file. For action-sensitive memories, record when the instruction applies, when it expires or what unlocks action, what to avoid, and who owns the instruction. OpenClaw memory documentation
Search or read a specific memory
memory_search finds relevant notes with semantic search, even when your query wording differs from the note. With an embedding provider configured, it combines semantic vector similarity with exact-term keyword matching. memory_get reads a specific file or line range. OpenClaw memory documentation
Where KnowMan fits
KnowMan saves articles and links, reads and tags them, answers questions with citations, and keeps memories such as claims, decisions, and playbooks. Its MCP server runs locally with npx -y @knowman/mcp and as a hosted server at https://api.knowman.io/mcp with OAuth sign-in or an API key. KnowMan MCP guide
KnowMan documents tools named kb_search, kb_get_item, kb_ask, and kb_save. KnowMan MCP guide See a guide to knowledge bases for AI agents for more on personal knowledge bases.
Connect OpenClaw to the hosted KnowMan server
OpenClaw can use KnowMan without a local install by pointing a remote MCP server at https://api.knowman.io/mcp. In clients that support remote MCP servers you add the server URL and sign in: OAuth opens a KnowMan consent page where you approve the knowman.read and knowman.write scopes, and access tokens last one hour and refresh automatically.
The other route is an API key: create one in your KnowMan settings and send it as an Authorization: Bearer <key> header. If you prefer to stay local, the open-source client at https://github.com/Essatt/knowman-mcp runs with npx -y @knowman/mcp.
Questions people ask
Where should I put stable instructions for OpenClaw?
Use USER.md for stable preferences, communication style, relationships, and active-project context. Write these as imperative directives. OpenClaw memory documentation
Does OpenClaw remember everything from daily notes automatically?
Daily notes are indexed for memory_search and memory_get, but are not injected into the bootstrap prompt on every turn. Today's and yesterday's dated notes load automatically after a bare /new or /reset. OpenClaw memory documentation
Can KnowMan connect to OpenClaw?
KnowMan has a local stdio MCP server. KnowMan MCP guide OpenClaw's memory documentation describes its own Markdown memory files. OpenClaw memory documentation
Does memory enforce approval rules?
No. OpenClaw's documentation says memory can preserve approval context but does not enforce policy. It points to approval settings, sandboxing, and scheduled tasks for hard operational controls. OpenClaw memory documentation