Search

Search

Search the published model library, indexed news, AI research papers, model developers and inference providers with one keyword query. Results are grouped by category, with a separate exact match count for each group.

Use GET /api/v1/search?q=Claude. Omitting category returns all five groups; add category=papers to page through just the matching papers. limit applies to each returned group, not to the combined response. Use the returned group's offset, limit and has_more to navigate. The website previews six results per group and shows 18 per page when a category is selected.

Keyword matching

This prototype uses PostgreSQL full-text search with the simple configuration. Search is case-insensitive and matches indexed words. Multiple unquoted words are combined with AND; double quotes request a phrase, OR accepts either alternative, and a leading minus excludes a word. Model and organization names, model aliases and titles receive greater weight than descriptions and article content. Results are ordered by relevance with a stable ID tie-breaker, independently within each category.

This is keyword search, not semantic search: a question does not imply synonyms, translations or an embedding query. The built-in parser does not provide Chinese word segmentation or fuzzy spelling correction. Ordinary punctuation follows PostgreSQL's tokenization rules. The query is limited to 200 Unicode characters; a blank query returns empty groups without reading the index. Expressions without an indexable positive term, including minus-only searches and an OR branch that matches everything except a word, return no matches rather than scanning the entire corpus. Account content, Operations data and unpublished source records are outside this endpoint.

Results and freshness

Each hit includes id, title, url, and nullable description, image_url and subtitle. A news URL points to the original article, a paper URL to its source publication, and model/developer/provider URLs to TensorTrace pages. Descriptions are plain text; clients must escape them before rendering. Nullable values indicate unavailable data.

The search index is maintained transactionally when public read data changes. No external publisher, model provider or AI service is contacted by a search request. Each response reflects committed database contents; offset pages may change as workers publish updates. Embeddings and semantic ranking are deferred to a later iteration.

The Search page provides the same grouped results in server-rendered HTML, with JavaScript enhancements for navigation. The API contract below is shared with other clients.

On this page

Search public AI content by keyword

GET /api/v1/search

PostgreSQL full-text keyword search over the public read index. Results and counts are grouped by category. No semantic model or remote acquisition runs during a request.

Access: public read; no TensorTrace API key.

Parameters

NameLocationRequiredType / limitsMeaning
qqueryNostring Default: ""; Max. length: 200Keyword query, trimmed before searching. Blank returns empty groups. Uses case-insensitive simple PostgreSQL web-search syntax: unquoted terms are AND, quoted text is a phrase, OR accepts either alternative, and minus excludes a term.
categoryqueryNostring Values: models, news, papers, developers, providersReturn only this category; omit to return all five groups in the documented order.
limitqueryNointeger Default: 6; Minimum: 1; Maximum: 18Maximum hits per returned category, independently of other groups.
offsetqueryNointeger Default: 0; Minimum: 0; Maximum: 10000Number of ranked hits to skip in each selected category.

Request example

curl 'https://tensortrace.net/api/v1/search?limit=1'

Responses

StatusMeaningBody type
200Grouped matches from committed public data.UnifiedSearchResponse
400Invalid query, category or pagination bounds.ErrorResponse
500A database read could not complete.ErrorResponse
503The public database is temporarily busy; retry shortly.ErrorResponse

Response example · 200

Illustrative values; schema-valid and not a live catalog measurement.

{
    "query": "example",
    "total": 1,
    "groups": [
        {
            "category": "models",
            "total": 1,
            "offset": 0,
            "limit": 1,
            "has_more": false,
            "items": [
                {
                    "id": "google/gemini-2.5-flash",
                    "title": "example",
                    "url": "example",
                    "description": "Illustrative model observation; consult live evidence for current specifications.",
                    "image_url": "example",
                    "subtitle": "example"
                }
            ]
        }
    ]
}
FieldTypeRequiredMeaning / constraints
querystringYesNormalized query used for matching.
totalintegerYesSum of match counts for the returned groups. Minimum: 0
groupsarray of UnifiedSearchGroupYesModels, news, papers, developers and providers in that order, or one selected category.

Models

Loading
EVIDENCE / COMPARISON

Benchmarks

Explore every benchmark

Loading

These charts contain model-linked results. The charts above also include unlinked source profiles. Compare scores within each evaluation. Model quality, usage, transparency and system performance measure different things.

Install TensorTrace