Alphaxiv Discover Papers
Discovers and ranks multiple candidate papers for a research topic. Use this for literature discovery, related work, broad topical coverage, or what is newest on a topic. Do not use it to locate one known paper — name that paper to a PDF tool directly. You get 2 searches per message, so cover every facet of the request in one broad call rather than spending them on several narrow ones. Provide request_body with an English question, focused keywords drawn from the user's request or previous tool results, and a difficulty estimate from 1 to 10 for retrieval effort. Translate non-English keyword terms into English, but preserve acronyms and names rather than guessing their expansions. Higher difficulty takes longer and enables multi-round searches. Optionally set prioritize to default, recency, historical, or popular to emphasize recent-year work, newest matches, foundational work, or community votes, respectively, after topical relevance. Use published_after and published_before (YYYY-MM-DD, inclusive) only when an explicit publication-date boundary is needed; to favor newer work without excluding older papers, use recency instead.
Pricing
| Condition | Charge |
|---|---|
| Each call | 1 credits/call |
Parameters
Optional parameters use the platform defaults when omitted.
| Parameter | Type | Required | Description | Default |
|---|---|---|---|---|
keywords | array | Yes | Only terms stated by the user or seen in other tool results, spelled as they appear. For "compare grpo and dapo" the keywords are ["GRPO", "DAPO"] and nothing more — never a guessed expansion like "group relative policy optimization", even when confident in it: short names collide, and the expansion buries the sense actually asked about. Focused terms are all you need; general-purpose terms, unlike most keyword searches, reduce result quality. The index is in English, so a term written in another language goes in translated ("具身智能" -> "embodied intelligence") while acronyms and names stay as written. | — |
question | string | Yes | What the person is asking for, in their terms, written in English. Length is not the goal here; a short query in their words beats a padded one in yours. For example, do not add the extra clause "in the context of" unless the prior tool results indicate such. | — |
difficulty | number | Yes | A 1-10 estimate of how much retrieval effort this request warrants. Higher values take longer, but trigger multi-round searches. | — |
prioritize | enum (historical | default | recency | popular) | No | What to weigh once topical relevance is accounted for. Omit it for the default, which grounds results in roughly the past year of work while letting strong older matches through. Use 'recency' when the request is about what is new — 'latest work', 'what has X been up to', 'recent papers' — which ranks the newest plausible matches first, needing no date bound and excluding nothing. Use 'historical' for seminal, foundational, or best-known work, where lifetime standing outweighs freshness. Use 'popular' only where the request names popularity, votes, or community standing — 'best' on its own asks which work is strongest, not which is most upvoted — which gives community votes the dominant weight among the topically relevant matches and mostly ignores age; for a pure most-popular list with no subject to search for, browse_paper_feed is the better tool. | default |
published_after | string | No | Only return papers first published on or after this date (YYYY-MM-DD). Omit it unless the request names a real boundary, e.g. 'since 2024' — this excludes papers outright, so to merely favour new work set prioritize to 'recency' instead of guessing a cutoff. | — |
published_before | string | No | Only return papers first published on or before this date (YYYY-MM-DD). Omit it unless the request names a historical window, e.g. 'before AlexNet' or 'papers from 2025'. Results thin out the further back and narrower the window is; report what comes back rather than retrying the same window. | — |
Invocation example
Parameter templateGenerated from the parameter configuration. Replace the sample values before use.
/agent-api/v1/tool/executeX-Api-Keycurl --request POST \
--url 'https://agentearth.ai/agent-api/v1/tool/execute?tool_name=ae_alphaxiv_discover_papers' \
--header "X-Api-Key: $AGENTEARTH_API_KEY" \
--header 'Content-Type: application/json' \
--data-raw '{
"params": {
"keywords": [
"xxx"
],
"question": "xxx",
"difficulty": 1
}
}'No response example available yet.




