Agent Memory
How WorkClaw agents remember context across sessions using MEMORY.md, embeddings, and daily logs for persistent, cross-session knowledge.
How does agent memory work?
Your Claw has three layers of memory that work together to maintain context across sessions:
- MEMORY.md — A structured text file where the agent stores important facts, preferences, and ongoing context.
- Embeddings — A semantic search index that lets the agent recall relevant information from past interactions.
- Daily logs — Automatic summaries of each day's activity that the agent can reference.
What is MEMORY.md?
MEMORY.md is the most visible layer of your Claw's memory. It is a plain-text markdown file that lives in your workspace's file system. Your Claw writes to it whenever it encounters information worth remembering — your preferences, project details, key decisions, recurring contacts.
You can view and edit MEMORY.md directly. This gives you full control over what your Claw knows. If something is wrong or outdated, just delete the line. If there is something important the agent should always know, add it yourself.
How do embeddings work?
Behind the scenes, WorkClaw indexes your conversations and files into an embedding system — a semantic search layer. When your Claw needs context it does not have in the current session, it searches this index to find relevant past conversations, documents, and notes.
This happens automatically. You do not need to tag, label, or organize anything. The embedding system understands meaning, so your Claw can find relevant context even when the exact words differ.
What are daily logs?
At the end of each day, WorkClaw generates a daily log summarizing what your Claw did — tasks completed, conversations had, actions taken. These logs are stored in your workspace and serve as a rolling record your agent can reference.
Daily logs help your Claw answer questions like "What did we discuss yesterday?" or "What tasks did I finish last week?" without needing to search through every individual session.
How does memory affect Chat sessions?
When you start a new Chat session, your Claw loads its MEMORY.md and can query its embeddings. This means it knows your preferences, ongoing projects, and recent activity — even though the conversation itself is fresh. The result is an agent that feels continuous, not forgetful.
For instructions and personality settings that should always apply, use Agent Configuration rather than relying on memory alone.