Skip to main content
The Memory Viewer shows you the files your OpenClaw agent has access to in its workspace memory directory (~/.openclaw/workspace*/). Use it to understand what context your agent is working with, spot oversized files before they inflate token costs, and verify that memory files contain the content you expect.

​
Agent tabs

If you run multiple OpenClaw agents, each agent gets its own tab at the top of the page. Click a tab to switch to that agent’s memory files. The number next to each agent name shows how many memory files it has.

​
File list

The left panel lists all files found in the agent’s memory directory. Each entry shows:
  • File name
  • Relative path within the workspace (shown below the name when the path differs from the filename)
  • Last modified time — shown as a time if modified today, or as a date otherwise
  • Estimated token count — an approximate token estimate for the file’s contents (see below)
The column header above the file list is labeled ctx cost as a reminder that every file in this list may be loaded into the agent’s context window. Click any file to open it in the content viewer on the right.

​
Content viewer

The right panel shows the full contents of the selected file as read-only text. The header bar displays:
  • The file’s relative path
  • The file’s absolute path on disk (abbreviated to ~ for your home directory)
The content area renders the file as plain text with monospace formatting, suitable for reading markdown, JSON, YAML, or any other text format.
The Memory Viewer is read-only. You cannot edit files from the dashboard.

​
Token estimates

The token estimate shown in the file list is a rough approximation, not an exact count from the model’s tokenizer. Claw Lens uses the following heuristic:
  • If more than 30% of the file’s characters are CJK (Chinese, Japanese, or Korean), it estimates 2 characters per token.
  • Otherwise, it estimates 4 characters per token, which is typical for English prose and code.
The estimate is displayed as ~N tok to make the approximation explicit.
Use token estimates to catch unexpectedly large memory files early. A single file estimated at 50k tokens will take up a significant fraction of a 200k-token context window, leaving less room for conversation history and tool results.

​
Use cases

Verifying agent context — If your agent is behaving unexpectedly, check which files are in its memory directory. A stale or incorrect memory file may be providing misleading context. Monitoring memory growth — Agents that write to their own memory directory can accumulate large files over time. The token estimates let you spot files that have grown large enough to affect performance. Debugging file contents — When an agent reads a memory file as part of a tool call (visible in the Session Timeline), you can open the same file here to read exactly what the agent saw.

​
Empty state

If no files are found, Claw Lens shows a message pointing to the ~/.openclaw/workspace*/ directory. This means either the agent has not created any memory files yet, or the workspace directory is in a non-default location.