Skip to main content

​
Full-Text Search

Full-text search finds sessions containing specific words or phrases. It uses Convex’s built-in search indexes, so results are fast and ranked by relevance.

​
How it works

OpenSync maintains a search index on the sessions.searchableText field. This field is built by concatenating the session title with all message text content. When you search, Convex’s search engine tokenizes your query and matches against this index.

​
What gets indexed

SourceIndexed field
Session titlesessions.searchableText (prefix)
User messagessessions.searchableText (appended)
Assistant messagessessions.searchableText (appended)
Tool call names and arguments are not included in searchableText by default. Only the textual content of messages is indexed.

​
Ranking

Results are ranked by Convex’s built-in relevance scoring, which factors in:
  • Term frequency (how often the term appears in the session)
  • Inverse document frequency (terms that are rare across all sessions score higher)
  • Position (terms appearing earlier in the document score slightly higher)

​
Search syntax

​
Single term

authentication
Finds sessions containing “authentication” anywhere in the searchable text.

​
Multiple terms

react hooks useState
All terms must be present (AND logic). This finds sessions containing “react”, “hooks”, and “useState”.

​
Prefix matching

refact
Matches “refactoring”, “refactored”, “refactor”, etc. Convex search supports prefix matching by default. Exact phrase matching is not currently supported. Multi-word queries use AND logic across individual terms.

​
Using full-text search in the dashboard

The search box at the top of the Sessions, Evals, and Context views all use full-text search by default.
  1. Type your query in the search box.
  2. Results update in real time as you type.
  3. Results are sorted by relevance score.
You can combine search with filters (source, date range, eval status) to narrow results further.

​
Using full-text search via API

curl -X POST "https://polished-penguin-622.convex.site/search" \
  -H "Authorization: Bearer osk_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "authentication middleware express",
    "type": "fulltext",
    "limit": 10
  }'
Response: 
{
  "results": [
    {
      "sessionId": "j57a...",
      "title": "Express Auth Middleware Setup",
      "snippet": "...setting up authentication middleware for the Express app...",
      "score": 0.95
    },
    {
      "sessionId": "k82b...",
      "title": "Add JWT middleware",
      "snippet": "...authentication layer using JSON web tokens...",
      "score": 0.78
    }
  ]
}

​
Tips for better results

  • Use specific terms. “convex mutation upsert” will return more targeted results than “database update”.
  • Include technical identifiers. Function names, library names, error codes, and file paths make great search terms.
  • Combine with source filters. If you know which tool you used, filter by source to narrow the results.
  • Use fewer terms for broader results. Start with one or two key terms and add more if you get too many results.

​
When to use full-text vs semantic

Use full-text whenUse semantic when
You know the exact terms usedYou want to describe the concept
Searching for function/variable namesSearching for solutions to a problem
Looking for specific error messagesLooking for similar approaches
Need fast, precise matchingNeed meaning-based matching
For more on semantic search, see Semantic Search. For combining both, see Hybrid Search.