Skip to content

Document Indexing and Memory Search

Your ClawOps bot searches locally before it asks an AI model to answer. That keeps responses faster and reduces the amount of context sent to the model. Two search engines share the work:

  • Builtin memory search (part of OpenClaw) covers what your assistant knows and saves.
  • QMD (a standalone document search tool) covers named document collections.

The QMD side is visible from Dashboard -> Content -> Documents.

What gets searched

Section titled “What gets searched”

Builtin memory search covers:

  • Agent memory: MEMORY.md, daily logs, and context the assistant writes down.
  • Agent learnings: rules your assistant has learned from previous mistakes or preferences.
  • User files: markdown and text files saved in ~/files.
  • Previous conversations: searchable session transcripts.

QMD covers:

  • Companion collections: specialist data, such as Entertainment Buddy taste profiles.
  • Your own collections: folders you ask the bot to index.

The Documents panel shows the live QMD collection names returned by qmd status. Memory, learnings, and conversations do not appear there. You can read your assistant’s memory in the Memory tab.

The Documents panel

Section titled “The Documents panel”

The status card shows:

  • Collections: how many named QMD collections are registered, plus their names.
  • Documents: how many files are currently indexed.
  • Last Indexed: when the index was last refreshed.
  • Auto-indexing active: whether ClawOps can see recent indexing activity.

Use Update Now when you have just added files to a collection and want them searchable immediately.

How indexing works

Section titled “How indexing works”

Both engines have two layers:

  • Keyword index: fast text search for exact and close matches.
  • Embeddings: semantic search for meaning-based matches.

Builtin memory search keeps itself up to date as files and conversations change. QMD collections refresh when companions update their data, and Update Now runs the QMD update and embedding steps on your VPS on demand.

Add a collection

Section titled “Add a collection”

Ask your bot naturally:

Index all markdown files in ~/my-notes

Your assistant can register that folder as a QMD collection so future questions can search it. Markdown and text files work best because they are easy to quote, summarize, and reason over. You don’t need a collection for ~/files: builtin memory search already covers it.

Search indexed documents

Section titled “Search indexed documents”

You can ask broad questions:

Search my docs for information about AI models

Or ask direct questions:

What’s in my workspace about configuration?

Your assistant searches first, then sends the relevant snippets to the active AI model. This is more efficient than loading an entire folder or long document into context.

Files vs Documents

Section titled “Files vs Documents”

The Files tab is for browsing, previewing, downloading, and deleting files saved on your VPS. Markdown and text files in ~/files are picked up by builtin memory search automatically.

The Documents tab is for QMD collections. A file in a collection folder can exist before it has been indexed. If you need it searchable right away, click Update Now in Documents.

Privacy

Section titled “Privacy”

Both indexes live on your dedicated VPS, and search happens there first. When an AI model answers a question, it receives only the selected context your assistant sends along with the prompt.

To make semantic memory search work, short passages of memory and conversation text are sent to an embeddings model through your OpenRouter key, the same way your chat messages are. The index itself stays on your VPS.

Search reduces token usage and limits unnecessary context sharing, but it does not make model calls disappear. The model still receives the relevant snippets needed to answer.

Troubleshooting

Section titled “Troubleshooting”

If a document is not showing up in search:

  1. Confirm the file is markdown or text.
  2. Confirm it is somewhere that gets indexed: ~/files, or a QMD collection you added.
  3. For collections, click Update Now.
  4. Ask the bot to search for a specific phrase from the file.
  5. If search still misses it, run Repair & Update from the dashboard and try again.

While your bot is on the free Wanderer stack (credits under 10%), memory search uses keyword matching only. Exact phrases still work, and meaning-based matches come back once your credits are topped up.

For files that you only need to download or preview, use the file browser instead.