News
News and research are ingested by TensorTrace workers and stored in the same database as the model catalog. Public GET requests query those local tables; they do not contact an external news service or trigger acquisition.
Development preview may seed bounded public API snapshots using
scripts/mirror_preview_data.py: 200 news articles and 100 papers by default,
with --kind news|papers for a partial refresh. It uses these same local
query/write projections, with explicitly marked snapshot provenance and paused
acquisition. Snapshots include full public article text/media and paper metadata
with optional images, not publisher HTML or production worker/account state;
they do not establish crawler/worker acceptance. Publication/revision dates
retain their source precision. Missing paper previews stay empty and do not
mark pending HTML enrichment complete. The news-only compatibility command
scripts/mirror_preview_news.py retains its original behavior.
GET /api/v1/news defaults to AI and machine learning terms. Callers may set
language, keywords (a JSON array), operator=and|or, limit=1..100, and either
before or after. Defaults are English, OR and ten records. Use keywords=[]
for unfiltered news. Matching includes full article text; there is no second
relevance filter or article deduplication. Distinct records sharing a URL remain.
Feed payloads contain summaries, links and media without reading/transferring
full text. GET /api/v1/news/{id} returns the complete stored article text, including
original Markdown for native release notes.
summary is generated and stored by workers. The version-1 plain-text policy
joins body blocks, skips repeated title/byline and standalone dates, and
normalizes whitespace. Its 300-character limit includes a final … when
truncated. Markdown release bodies use version 2: parse the original structure
before clipping, flatten headings/lists/tables into compact prose, and preserve
balanced inline emphasis, code, safe links and complete TeX. Images and HTML
attributes do not become excerpt text. The preview contains at most 300 visible
Unicode characters including an optional trailing …; Markdown syntax and
link destinations can make the serialized string longer (bounded to 4 KiB).
No AI, remote requests or client-side summarization are involved. Topic/model
feeds and article details share this stored field. Original publisher excerpts,
complete Markdown bodies and acquisition checksums remain unchanged.
Existing ready plain articles are backfilled by the content workers. Retained
Markdown releases upgrade on the next publication/replay even when their
original body is unchanged; a second replay is a no-op. During plain-text
backfill, an original optional publisher excerpt remains the fallback.
GET /api/v1/models/{provider}/{model}/news uses the same query engine and
parameters. Its default terms come from the canonical model and developer,
including alias resolution; serving providers do not substitute for developers.
This entry remains useful for consistent model identities and installed native
clients. General searches use ordinary parameters; no separate relay or query
implementation exists.
Feed context describes effective terms. relevance is topic on topic feeds,
model or developer when excerpts mention those terms, otherwise related.
Labels describe excerpts and do not determine inclusion. Unsafe links/media are
sanitized at the display boundary; clients sanitize rich text and omit failed images.
News summaries/bodies and paper abstracts retain their original text in JSON.
Clients render Markdown and TeX ($…$, $$…$$, \(…\), \[…\]) at the
display boundary. Code and currency remain literal; source HTML stays inert.
Compact summaries clamp the rendered content, while expanded paper abstracts
render the complete stored abstract. Existing worker-generated 300-character
news summaries remain authoritative; malformed/truncated formulas fall back
to readable text. There are no rendered-HTML API fields or new acquisition
requests. See rich-text presentation.
Use pagination.older_cursor with before, or newer_cursor with after.
Cursors are bound to effective query filters. Matching/unfiltered totals and
paging flags come from the same database snapshot. Versioned continuation URLs
are also supplied. next_cursor and upstream_total remain aliases for installed
clients. published_at is the publisher timestamp or null; created_at is the
first receipt fallback. fetched_at is query completion time. Feed pages are
not cached; raw inputs and extracted articles retain their provenance.
source.icon_url is optional. image_urls keeps up to twenty safe original
preview URLs. Hacker News's source icon/name and discussion URL identify its
post; the headline URL points to the article. Images link to publisher media.
GET /api/v1/research returns source-neutral AI/ML research works and optional
previews, with ten records by default. GET /api/v1/papers remains a compatible
spelling for installed clients and existing links. Filter with category, JSON keywords,
operator=and|or, limit, and either before or after. Paper keywords match
literal titles/abstracts. Categories are cs.AI, cs.LG, cs.CL, cs.CV, cs.NE and
stat.ML. Paper cursors are bound to filters; a YYYY-MM-DD boundary is also valid.
Literal substring search retains exact counts and can match broad terms inside
longer words; its cost depends on the matching corpus. Use the unified keyword
search API for token-based relevance rather than treating substring counts as
a fixed-latency operation.
Use pagination.next_cursor/next for older results and
previous_cursor/previous for newer results.
Every work has a stable UUID research_id and canonical website
research_url (/research/{uuid}). The legacy integer id remains available
for installed clients, cursor ties and compatibility links. arxiv_id is
nullable: arXiv is one external identity, not the identity of all research.
GET /api/v1/research/{uuid-or-integer} and /api/v1/papers/{integer} return
the same work projection, including the abstract, authors, categories and
source-native URLs. url is the preferred external reading URL; pdf_url and
figures remain optional.
sources contains {source_id, name, url, external_id} records. Explicit
cross-references can place arXiv, Hugging Face Papers, DOI registries and
OpenReview on one work; the client does not match titles or merge identities.
A work with only one source is valid. HF Papers and configured public
OpenReview submissions can originate works. OpenAlex, Crossref and Semantic
Scholar currently enrich exact existing works; unmatched registry observations
remain retained until an indexed work provides an in-scope identity anchor. Different publications remain in
versions: {id, kind, label, source_id, url, published_on, revised_on}.
Corrections and retractions are notices with {kind, title, url, published_on}.
reviews keeps the source URL, venue, decision, review count and original
labelled ratings in the API. Public clients present supporting records as
concise external links, rather than review/audit metadata panels. Notice kinds
remain in their link labels, including notices sharing the work URL. No client
averages recommendation values into a quality score. references attaches
source-qualified dataset audits/schema and other supporting material to its
work; reported license/provenance assertions are not legal verification.
Source availability is independent: an active alternative can keep a work visible after another source withdraws it. Withdrawn/deleted observations stay in source provenance with their availability, but do not supply the current reading URL, versions, reviews, decisions or figures. Existing native arXiv visibility remains authoritative: a hidden native paper is not reactivated by an active auxiliary source. Restoration preserves the work UUID and aliases.
Publication dates preserve source precision: published_on is
YYYY-MM-DD, YYYY-MM or YYYY, or null when unknown. published_precision
describes that precision. Existing arXiv published_date preserves the numeric
source form YYYYMMDD or YYYYMM00. A month or year never implies publication
on day 01, midnight or a fabricated timestamp. Each version retains its own
publication/revision dates rather than borrowing the fetch time.
Opaque cursors use the internal calendar sort anchor and ID, so month-only
records and genuine first-of-month records paginate without missing ties.
image_urls is always an array: arXiv contributes its first usable original
figure; HF-only works can retain up to five source media URLs. Clients preview
the first image. An empty
array means not yet indexed, unavailable HTML, no usable figure or enrichment
failure; it does not mean the paper is unavailable. List and detail use the same
current metadata/figure projection. Images do not affect search, totals or paging.
Workers ingest source metadata and independently fetch the optional arXiv HTML reader to extract the first figure. HTTP 404/410 is allowed absence. HTML bytes and extraction provenance stay internal; there is no PDF download, full-text index or request-time acquisition. Clients access the original arXiv image. See figure architecture.
Invalid queries return 400; missing details return 404; database unavailability
returns 503. Errors use TensorTrace's {"error":"message"} contract. News and
paper availability do not affect model catalog reads.
Official release notes and package references
Formal GitHub releases from the configured projects can appear in the same News list/detail API as publisher articles. They retain the original release timestamp, official project attribution and complete retained release notes. Drafts, prereleases, empty notes and unknown publication dates do not become fresh News. Source records and immutable receipts remain available internally; an unchanged replay does not create another article or change its date.
references is an additive array, empty for articles without supporting package
metadata. A release may include up to five package_metadata references with
source_id, name, url, version, requires_python, requires_dist,
dependency_count, license, receipt_id, observed_at and scope.
requires_dist contains at most 50 declared dependency entries; the count
describes the complete source array. These are source declarations, not a
dependency resolution, compatibility guarantee or security assessment. Public
clients display the package name/version as an external link only; dependency
arrays, Python requirements, observation clocks and receipt IDs remain API data.
Only the same configured project and exact release/package version match
(an initial release-tag v is ignored). The llama-cpp-python binding is not
the upstream llama.cpp package identity. Package observations remain separate
from the official notes and publication clock. Clients can ignore the additive
array without changing existing article handling. An older release keeps the
matching retained package revision and its original receipt/observation clock
when the package's latest version advances.
On this page
- Read recent AI and machine learning news ·
GET /news - Read model and developer news ·
GET /models/{provider}/{model}/news - Read a complete stored news article ·
GET /news/{id} - Read source-neutral Research (compatible Papers route) ·
GET /papers - Read one stored research paper ·
GET /papers/{id} - Read source-neutral Research works ·
GET /research - Read Research by stable UUID or numeric alias ·
GET /research/{id}
Read recent AI and machine learning news
GET /api/v1/news
Read locally ingested first-party news. AI/model terms are defaults; callers may control language, keywords, operator, page size and either cursor. Feeds omit full text; detail reads expose it. Search includes article bodies and retains shared-URL records.
Access: public read; no TensorTrace API key.
Parameters
| Name | Location | Required | Type / limits | Meaning |
|---|---|---|---|---|
before | query | No | string Max. length: 2048 | Older cursor returned by this query, or a timestamp. Filters must match. |
after | query | No | string Max. length: 2048 | Newer cursor returned by this query, or a timestamp. Use either before or after. |
limit | query | No | integer Default: 10; Minimum: 1; Maximum: 100 | Maximum records in one page. |
language | query | No | string Default: "en"; Max. length: 63 | One language tag; regions are ignored and script variants remain selectable. |
keywords | query | No | string Max. length: 8192 | JSON array of at most 16 keyword entries. Words within an entry use AND; [] reads all news. Omit for AI topic/model defaults. |
operator | query | No | string Default: "or"; Values: and, or | How keyword entries combine; no second title/summary relevance filter. |
Request example
curl 'https://tensortrace.net/api/v1/news?limit=1'
Responses
| Status | Meaning | Body type |
|---|---|---|
200 | Current first-party database results; no relay or feed-page cache. | NewsFeed |
400 | Invalid query or cursor filters. | ErrorResponse |
503 | Database temporarily unavailable. | ErrorResponse |
Response example · 200
Illustrative values; schema-valid and not a live catalog measurement.
{
"data": [
{
"id": 1,
"title": "example",
"summary": "example",
"url": "https://example.com/model",
"source": {
"name": "Google: Gemini 2.5 Flash",
"url": "example"
},
"published_at": "2026-10-05T00:00:00Z",
"created_at": "2026-10-05T00:00:00Z",
"discussion_url": "https://example.com/model",
"relevance": "topic"
}
],
"context": {
"kind": "topic",
"model_id": "google/gemini-2.5-flash",
"developer": "example",
"model_terms": [
"example"
],
"developer_terms": [
"example"
],
"keywords": [
"example"
],
"operator": "and",
"language": "example"
},
"fetched_at": "2026-10-05T00:00:00Z",
"upstream_total": 1,
"source": "TensorTrace",
"next_cursor": "example",
"pagination": {
"language": "example",
"keywords": [
"example"
],
"operator": "and",
"total": 1,
"unfiltered_total": 1,
"count": 1,
"limit": 1,
"has_older": false,
"has_newer": false,
"older_cursor": "example",
"newer_cursor": "example",
"older_url": "example",
"newer_url": "example"
}
}
| Field | Type | Required | Meaning / constraints |
|---|---|---|---|
data | array of NewsArticle | Yes | Matching first-party articles in publication order; body-only and shared-URL records are retained. Max. items: 100 |
context | NewsContext | Yes | |
fetched_at | string · date-time | Yes | Time this database query completed; not source acquisition or publication time. |
upstream_total | integer | Yes | Matching database entry count retained for installed-client compatibility. Minimum: 0 |
source | string | Yes | First-party service. Values: TensorTrace |
next_cursor | string,null | Yes | Opaque continuation for older results, or null when exhausted. |
pagination | NewsPagination | Yes |
Read model and developer news
GET /api/v1/models/{provider}/{model}/news
Read locally ingested first-party news. AI/model terms are defaults; callers may control language, keywords, operator, page size and either cursor. Feeds omit full text; detail reads expose it. Search includes article bodies and retains shared-URL records.
Access: public read; no TensorTrace API key.
Parameters
| Name | Location | Required | Type / limits | Meaning |
|---|---|---|---|---|
provider | path | Yes | string | Canonical model provider segment; aliases resolve to current identity. |
model | path | Yes | string | Canonical model model segment; aliases resolve to current identity. |
before | query | No | string Max. length: 2048 | Older cursor returned by this query, or a timestamp. Filters must match. |
after | query | No | string Max. length: 2048 | Newer cursor returned by this query, or a timestamp. Use either before or after. |
limit | query | No | integer Default: 10; Minimum: 1; Maximum: 100 | Maximum records in one page. |
language | query | No | string Default: "en"; Max. length: 63 | One language tag; regions are ignored and script variants remain selectable. |
keywords | query | No | string Max. length: 8192 | JSON array of at most 16 keyword entries. Words within an entry use AND; [] reads all news. Omit for AI topic/model defaults. |
operator | query | No | string Default: "or"; Values: and, or | How keyword entries combine; no second title/summary relevance filter. |
Request example
curl 'https://tensortrace.net/api/v1/models/google/gemini-2.5-flash/news?limit=1'
Responses
| Status | Meaning | Body type |
|---|---|---|
200 | Current first-party database results; no relay or feed-page cache. | NewsFeed |
400 | Invalid query or cursor filters. | ErrorResponse |
404 | Model not found. | ErrorResponse |
503 | Database temporarily unavailable. | ErrorResponse |
Response example · 200
Illustrative values; schema-valid and not a live catalog measurement.
{
"data": [
{
"id": 1,
"title": "example",
"summary": "example",
"url": "https://example.com/model",
"source": {
"name": "Google: Gemini 2.5 Flash",
"url": "example"
},
"published_at": "2026-10-05T00:00:00Z",
"created_at": "2026-10-05T00:00:00Z",
"discussion_url": "https://example.com/model",
"relevance": "topic"
}
],
"context": {
"kind": "topic",
"model_id": "google/gemini-2.5-flash",
"developer": "example",
"model_terms": [
"example"
],
"developer_terms": [
"example"
],
"keywords": [
"example"
],
"operator": "and",
"language": "example"
},
"fetched_at": "2026-10-05T00:00:00Z",
"upstream_total": 1,
"source": "TensorTrace",
"next_cursor": "example",
"pagination": {
"language": "example",
"keywords": [
"example"
],
"operator": "and",
"total": 1,
"unfiltered_total": 1,
"count": 1,
"limit": 1,
"has_older": false,
"has_newer": false,
"older_cursor": "example",
"newer_cursor": "example",
"older_url": "example",
"newer_url": "example"
}
}
| Field | Type | Required | Meaning / constraints |
|---|---|---|---|
data | array of NewsArticle | Yes | Matching first-party articles in publication order; body-only and shared-URL records are retained. Max. items: 100 |
context | NewsContext | Yes | |
fetched_at | string · date-time | Yes | Time this database query completed; not source acquisition or publication time. |
upstream_total | integer | Yes | Matching database entry count retained for installed-client compatibility. Minimum: 0 |
source | string | Yes | First-party service. Values: TensorTrace |
next_cursor | string,null | Yes | Opaque continuation for older results, or null when exhausted. |
pagination | NewsPagination | Yes |
Read a complete stored news article
GET /api/v1/news/{id}
Access: public read; no TensorTrace API key.
Parameters
| Name | Location | Required | Type / limits | Meaning |
|---|---|---|---|---|
id | path | Yes | integer Minimum: 1 | Stored first-party numeric record ID returned by the list. |
language | query | No | string Default: "en"; Max. length: 63 | One language tag; regions are ignored and script variants remain selectable. |
Request example
curl 'https://tensortrace.net/api/v1/news/google'
Responses
| Status | Meaning | Body type |
|---|---|---|
200 | First-party database response. | NewsDetail |
400 | Invalid request or resource unavailable. | ErrorResponse |
404 | Invalid request or resource unavailable. | ErrorResponse |
503 | Invalid request or resource unavailable. | ErrorResponse |
Response example · 200
Illustrative values; schema-valid and not a live catalog measurement.
{
"id": 1,
"title": "example",
"summary": "example",
"url": "https://example.com/model",
"source": {
"name": "Google: Gemini 2.5 Flash",
"url": "example"
},
"published_at": "2026-10-05T00:00:00Z",
"created_at": "2026-10-05T00:00:00Z",
"content": "example",
"language": "example",
"author": "example"
}
| Field | Type | Required | Meaning / constraints |
|---|---|---|---|
id | integer | Yes | Selected TensorTrace news-entry ID. |
title | string | Yes | Original article title. |
summary | string,null | Yes | Stored description from accepted article/release text, at most 300 characters with a final ellipsis when shortened. |
url | string · uri | Yes | Selected original HTTP(S) article URL. |
source | NewsPublisher | Yes | |
published_at | string,null · date-time | Yes | Source publication time, null when missing. |
created_at | string · date-time | Yes | TensorTrace receipt time and ordering fallback. |
discussion_url | string,null · uri | No | Optional original discussion URL. |
image_urls | array of string · uri | No | Safe, deduplicated original article images in TensorTrace order, preview first. Empty when unavailable; clients may show the first image. Max. items: 20 |
content | string | Yes | Complete plain-text article; no raw extraction HTML. |
language | string | Yes | |
author | string,null | Yes | |
references | array of PackageReference | No | Optional exact-version package evidence, separate from the article/release text. Max. items: 5 |
Read source-neutral Research (compatible Papers route)
GET /api/v1/papers
Access: public read; no TensorTrace API key.
Parameters
| Name | Location | Required | Type / limits | Meaning |
|---|---|---|---|---|
before | query | No | string Max. length: 2048 | YYYY-MM-DD boundary or returned opaque cursor bound to this query. |
after | query | No | string Max. length: 2048 | YYYY-MM-DD boundary or returned opaque cursor bound to this query. |
limit | query | No | integer Default: 10; Minimum: 1; Maximum: 100 | Maximum records in one page. |
keywords | query | No | string Max. length: 8192 | JSON array of at most 16 literal entries matched in title and abstract. No full PDF retrieval. |
operator | query | No | string Default: "and"; Values: and, or | How keyword entries combine; no second title/summary relevance filter. |
category | query | No | string Values: cs.AI, cs.LG, cs.CL, cs.CV, cs.NE, stat.ML | Filter by one supported arXiv AI/ML category. |
Request example
curl 'https://tensortrace.net/api/v1/papers?limit=1'
Responses
| Status | Meaning | Body type |
|---|---|---|
200 | First-party database response. | PaperPage |
400 | Invalid request or resource unavailable. | ErrorResponse |
404 | Invalid request or resource unavailable. | ErrorResponse |
503 | Invalid request or resource unavailable. | ErrorResponse |
Response example · 200
Illustrative values; schema-valid and not a live catalog measurement.
{
"data": [
{
"arxiv_id": "example",
"title": "example",
"abstract": "example",
"url": "example",
"id": 1,
"pdf_url": "example",
"doi": "example",
"license_url": "example",
"comments": "example",
"journal_ref": "example",
"authors": [
{}
],
"categories": [
"example"
],
"published_on": "2026-10-23",
"revised_on": "2026-10-23",
"created_at": "2026-10-05T00:00:00Z",
"updated_at": "2026-10-05T00:00:00Z",
"source": {
"name": "Google: Gemini 2.5 Flash",
"url": "example"
},
"image_urls": [
"https://example.com/model"
],
"published_date": 1,
"published_precision": "day"
}
],
"pagination": {
"count": 1,
"total": 1,
"unfiltered_total": 1,
"limit": 1,
"keywords": [
"example"
],
"operator": "and",
"category": "example",
"before": "example",
"after": "example",
"next_cursor": "example",
"previous_cursor": "example",
"next": "example",
"previous": "example"
}
}
| Field | Type | Required | Meaning / constraints |
|---|---|---|---|
data | array of Paper | Yes | Max. items: 100 |
pagination | PaperPagination | Yes |
Read one stored research paper
GET /api/v1/papers/{id}
Access: public read; no TensorTrace API key.
Parameters
| Name | Location | Required | Type / limits | Meaning |
|---|---|---|---|---|
id | path | Yes | string Pattern: ^(?:[1-9][0-9]*|[0-9a-fA-F]{8}-(?:[0-9a-fA-F]{4}-){3}[0-9a-fA-F]{12})$ | Stable Research UUID, or the existing positive numeric ID alias. |
Request example
curl 'https://tensortrace.net/api/v1/papers/google'
Responses
| Status | Meaning | Body type |
|---|---|---|
200 | First-party database response. | Paper |
400 | Invalid request or resource unavailable. | ErrorResponse |
404 | Invalid request or resource unavailable. | ErrorResponse |
503 | Invalid request or resource unavailable. | ErrorResponse |
Response example · 200
Illustrative values; schema-valid and not a live catalog measurement.
{
"arxiv_id": "example",
"title": "example",
"abstract": "example",
"url": "example",
"id": 1,
"pdf_url": "example",
"doi": "example",
"license_url": "example",
"comments": "example",
"journal_ref": "example",
"authors": [
{}
],
"categories": [
"example"
],
"published_on": "2026-10-23",
"revised_on": "2026-10-23",
"created_at": "2026-10-05T00:00:00Z",
"updated_at": "2026-10-05T00:00:00Z",
"source": {
"name": "Google: Gemini 2.5 Flash",
"url": "example"
},
"image_urls": [
"https://example.com/model"
],
"published_date": 1,
"published_precision": "day"
}
| Field | Type | Required | Meaning / constraints |
|---|---|---|---|
arxiv_id | string,null | Yes | External arXiv identity; null for works indexed only from other sources. |
title | string | Yes | |
abstract | string | Yes | |
url | string | Yes | |
id | integer | Yes | Minimum: 1 |
pdf_url | string,null | Yes | |
doi | string,null | Yes | |
license_url | string,null | Yes | |
comments | string,null | Yes | |
journal_ref | string,null | Yes | |
authors | array of object | Yes | |
categories | array of string | Yes | |
published_on | ResearchCalendarDate | Yes | |
revised_on | ResearchCalendarDate | Yes | |
created_at | string · date-time | Yes | |
updated_at | string · date-time | Yes | |
source | object | Yes | |
image_urls | array of string · uri | Yes | Up to 5 safe source media URLs. arXiv HTML enrichment supplies its first figure; HF metadata can supply more. Empty means unavailable or not yet indexed. Max. items: 5 |
ecosystem_url | string | No | Optional same-origin discovery link, supplied only when an exact arXiv/DOI relationship is indexed. |
published_date | integer,null | Yes | Compatible YYYYMMDD calendar integer; DD=00 for month precision, MMDD=0000 for year precision, null when unknown. Never a Unix timestamp. |
published_precision | string | Yes | Values: day, month, year, unknown |
research_id | string · uuid | No | Stable source-neutral work UUID. The positive numeric id remains a compatible alias. |
research_url | string | No | Same-origin canonical work page. |
sources | array of ResearchSource | No | Explicit external identities for one work; no title similarity matching. |
versions | array of ResearchVersion | No | Source-specific preprint/publication versions retain their own dates. |
notices | array of object | No | Correction/retraction relationships, not aliases or replacement publication dates. |
reviews | array of ResearchReview | No | |
references | array of ContentReference | No | Up to 3 compact list references or 20 full detail references; reference_count reports the complete total. Max. items: 20 |
reference_count | integer | No | All current exact supporting relationships; may exceed the bounded preview. Minimum: 0 |
availability | string | No | Stored work availability. Withdrawn/retracted evidence remains attributable. |
Read source-neutral Research works
GET /api/v1/research
One work, multiple explicitly linked sources and publication versions. Existing /papers routes and numeric IDs remain compatible aliases.
Access: public read; no TensorTrace API key.
Parameters
| Name | Location | Required | Type / limits | Meaning |
|---|---|---|---|---|
before | query | No | string Max. length: 2048 | YYYY-MM-DD boundary or returned opaque cursor bound to this query. |
after | query | No | string Max. length: 2048 | YYYY-MM-DD boundary or returned opaque cursor bound to this query. |
limit | query | No | integer Default: 10; Minimum: 1; Maximum: 100 | Maximum records in one page. |
keywords | query | No | string Max. length: 8192 | JSON array of at most 16 literal entries matched in title and abstract. No full PDF retrieval. |
operator | query | No | string Default: "and"; Values: and, or | How keyword entries combine; no second title/summary relevance filter. |
category | query | No | string Values: cs.AI, cs.LG, cs.CL, cs.CV, cs.NE, stat.ML | Filter by one supported arXiv AI/ML category. |
Request example
curl 'https://tensortrace.net/api/v1/research?limit=1'
Responses
| Status | Meaning | Body type |
|---|---|---|
200 | First-party database response. | PaperPage |
400 | Invalid request or resource unavailable. | ErrorResponse |
404 | Invalid request or resource unavailable. | ErrorResponse |
503 | Invalid request or resource unavailable. | ErrorResponse |
Response example · 200
Illustrative values; schema-valid and not a live catalog measurement.
{
"data": [
{
"arxiv_id": "example",
"title": "example",
"abstract": "example",
"url": "example",
"id": 1,
"pdf_url": "example",
"doi": "example",
"license_url": "example",
"comments": "example",
"journal_ref": "example",
"authors": [
{}
],
"categories": [
"example"
],
"published_on": "2026-10-23",
"revised_on": "2026-10-23",
"created_at": "2026-10-05T00:00:00Z",
"updated_at": "2026-10-05T00:00:00Z",
"source": {
"name": "Google: Gemini 2.5 Flash",
"url": "example"
},
"image_urls": [
"https://example.com/model"
],
"published_date": 1,
"published_precision": "day"
}
],
"pagination": {
"count": 1,
"total": 1,
"unfiltered_total": 1,
"limit": 1,
"keywords": [
"example"
],
"operator": "and",
"category": "example",
"before": "example",
"after": "example",
"next_cursor": "example",
"previous_cursor": "example",
"next": "example",
"previous": "example"
}
}
| Field | Type | Required | Meaning / constraints |
|---|---|---|---|
data | array of Paper | Yes | Max. items: 100 |
pagination | PaperPagination | Yes |
Read Research by stable UUID or numeric alias
GET /api/v1/research/{id}
One work, multiple explicitly linked sources and publication versions. Existing /papers routes and numeric IDs remain compatible aliases.
Access: public read; no TensorTrace API key.
Parameters
| Name | Location | Required | Type / limits | Meaning |
|---|---|---|---|---|
id | path | Yes | string Pattern: ^(?:[1-9][0-9]*|[0-9a-fA-F]{8}-(?:[0-9a-fA-F]{4}-){3}[0-9a-fA-F]{12})$ | Stable Research UUID, or the existing positive numeric ID alias. |
Request example
curl 'https://tensortrace.net/api/v1/research/google'
Responses
| Status | Meaning | Body type |
|---|---|---|
200 | First-party database response. | Paper |
400 | Invalid request or resource unavailable. | ErrorResponse |
404 | Invalid request or resource unavailable. | ErrorResponse |
503 | Invalid request or resource unavailable. | ErrorResponse |
Response example · 200
Illustrative values; schema-valid and not a live catalog measurement.
{
"arxiv_id": "example",
"title": "example",
"abstract": "example",
"url": "example",
"id": 1,
"pdf_url": "example",
"doi": "example",
"license_url": "example",
"comments": "example",
"journal_ref": "example",
"authors": [
{}
],
"categories": [
"example"
],
"published_on": "2026-10-23",
"revised_on": "2026-10-23",
"created_at": "2026-10-05T00:00:00Z",
"updated_at": "2026-10-05T00:00:00Z",
"source": {
"name": "Google: Gemini 2.5 Flash",
"url": "example"
},
"image_urls": [
"https://example.com/model"
],
"published_date": 1,
"published_precision": "day"
}
| Field | Type | Required | Meaning / constraints |
|---|---|---|---|
arxiv_id | string,null | Yes | External arXiv identity; null for works indexed only from other sources. |
title | string | Yes | |
abstract | string | Yes | |
url | string | Yes | |
id | integer | Yes | Minimum: 1 |
pdf_url | string,null | Yes | |
doi | string,null | Yes | |
license_url | string,null | Yes | |
comments | string,null | Yes | |
journal_ref | string,null | Yes | |
authors | array of object | Yes | |
categories | array of string | Yes | |
published_on | ResearchCalendarDate | Yes | |
revised_on | ResearchCalendarDate | Yes | |
created_at | string · date-time | Yes | |
updated_at | string · date-time | Yes | |
source | object | Yes | |
image_urls | array of string · uri | Yes | Up to 5 safe source media URLs. arXiv HTML enrichment supplies its first figure; HF metadata can supply more. Empty means unavailable or not yet indexed. Max. items: 5 |
ecosystem_url | string | No | Optional same-origin discovery link, supplied only when an exact arXiv/DOI relationship is indexed. |
published_date | integer,null | Yes | Compatible YYYYMMDD calendar integer; DD=00 for month precision, MMDD=0000 for year precision, null when unknown. Never a Unix timestamp. |
published_precision | string | Yes | Values: day, month, year, unknown |
research_id | string · uuid | No | Stable source-neutral work UUID. The positive numeric id remains a compatible alias. |
research_url | string | No | Same-origin canonical work page. |
sources | array of ResearchSource | No | Explicit external identities for one work; no title similarity matching. |
versions | array of ResearchVersion | No | Source-specific preprint/publication versions retain their own dates. |
notices | array of object | No | Correction/retraction relationships, not aliases or replacement publication dates. |
reviews | array of ResearchReview | No | |
references | array of ContentReference | No | Up to 3 compact list references or 20 full detail references; reference_count reports the complete total. Max. items: 20 |
reference_count | integer | No | All current exact supporting relationships; may exceed the bounded preview. Minimum: 0 |
availability | string | No | Stored work availability. Withdrawn/retracted evidence remains attributable. |