Search and Indexing

Common search and indexing problems and how to resolve them.

Search returns no results

Causes:

  • No project is currently open
  • Project files have not been indexed yet
  • Your query does not match any indexed content
  • You are searching the wrong project

Solutions:

  1. Make sure you have a project open in ThinkForge (search needs an active project).
  2. Check that your project has indexed content — look for indexed document counts in the project memory info.
  3. Try a simpler or shorter query.
  4. Try keyword search if semantic search returns nothing — your content may not be embedded yet.

Semantic search gives poor results but keyword search works

Symptom: Keyword matches are fine, but meaning-based search returns irrelevant results or nothing.

Cause: The local embedding model is missing or failed to initialize. Without it, semantic search is disabled and hybrid search falls back to keyword-only.

Solution:

  1. Check if the model files exist at Resources\Models\all-MiniLM-L6-v2\ in your ThinkForge installation (you need model.onnx and vocab.txt).
  2. If missing, reinstall or repair ThinkForge to restore the model files.
  3. After restoring, run an embedding backfill to generate vectors for existing content.

Files are not getting indexed

Symptom: You added files to your project but they do not appear in search results.

Causes:

  • The file type is not automatically indexed (only .md, .txt, .rtf, .json, .html, .css, .js, .cs, .py, .xml are auto-indexed from disk)
  • The file is in a hidden or excluded folder (.fire, .git, bin, obj, node_modules)
  • The file is deeper than 5 folder levels
  • The file was added while ThinkForge was closed (reconciliation runs on next project open)

Solutions:

  • For supported file types, close and reopen the project to trigger reconciliation.
  • For files in unsupported formats, use the "Add to Memory" feature from the browser extension or manually index through the memory tools.
  • Move deeply nested files to a shallower folder level if needed.

Document shows "No content indexed"

Symptom: A document appears in the Project Navigator but the info panel says no content is indexed.

Cause: The file is in the project folder but has not been processed by the indexing pipeline. This can happen with newly added files or unsupported file types.

Solution: Open the file in ThinkForge (which triggers indexing), or use the index/reconcile tools to process pending files.

Browser captures not appearing in search

Symptom: Content captured from the browser extension does not show up in search.

Cause: No active project was open in ThinkForge Desktop when the capture was made.

Solution: Always have a project open before using browser capture features. If content was captured without a project, open the target project and re-index or re-capture.

Cluster Graph is empty

Symptom: The Cluster Graph shows no nodes.

Cause: Documents need to be indexed with embeddings to appear in the cluster visualization. If embedding is disabled or no content has been indexed, the graph will be empty.

Solution: Ensure the embedding model is present and working, then index your project content. Run an embedding backfill if you have indexed documents without vectors.

Index errors

Symptom: Indexing fails with an error message.

Causes:

  • Disk space full — the memory database cannot grow
  • Write permissions issue on the project's .fire/memory folder
  • Corrupted database file

Solutions:

  • Free up disk space
  • Ensure your user account has write access to the project folder
  • If the database is corrupted, delete the memory database file and re-index (this rebuilds the index from your files)
Re-indexing is safe. The memory database is built from your project files. Deleting and rebuilding it does not lose any source content — only the search index is regenerated.

Related