Lemon8 Public Data API | SandBase
Read public Lemon8 search, posts, users, and topics with one REST API. No Lemon8 login, no SDK — one SandBase key, built for agent workflows.

Lemon8’s public catalog includes search, posts, users and topics, but those capabilities do not share one business parser. A developer starting with a search phrase needs a different input and validation path from one who already has a post ID. The useful first step is to choose one of those tasks and inspect a single response before building a larger discovery workflow.
This 2026-11-08 repair uses free inspect and public references for search and post-detail without making a new business call. The original 2026-09-27 UTC search sample remains historical evidence. User, topic and discover routes below are catalog navigation, not a freshly validated end-to-end workflow.
Key takeaway
- Use
queryfor the selected search, with explicit main tab and empty max_cursor for the first page.- The current main-tab reference is an object with
data.items[].articles[]; the old list-of-blocks sample is historical, not a fallback parser.- Post detail takes string
item_idand uses the saved-reference message success/data object checks.- A share link needs a separate resolver decision: current
lemon8/app/item-idtakesshare_text, not the old get_item_id route name.
The retained cover is conceptual artwork. Its creator names, content cards, counts and comment labels are illustrative, not current API measurements or verified identities.
Choose the capability for your input
| Input or need | Destination | Scope here |
|---|---|---|
| Search phrase | Lemon8 search | One main-tab page |
| Known numeric post ID | Post detail | One post lookup |
| Share text or link | Item ID resolver | Current route/schema identified; not executed |
| Posting or account-authorized actions | Lemon8 official channels | Outside this public read-only guide |
The selected routes use a SandBase key without a Lemon8 login in the request. That does not grant account authority or access to private content. SandBase supplies an access layer and does not own or operate the underlying platform.
The content discovery tutorial provides the checked Python helper and distinguishes an existing key, a connected agent and a new project. It does not automatically select the first post, fetch its author or enumerate every result.
Historical catalog screenshot retained as navigation. Endpoint count, labels and status in this image are not a fresh availability or quota measurement.
Quick start with one main-tab search
Keep SANDBASE_API_KEY in your local runtime or secret manager. Send query, main and an empty string cursor explicitly. This curl prints the response and sets time limits; the tutorial handles the business checks. It includes no retry or redirect-following option.
curl --fail-with-body --connect-timeout 10 --max-time 90 \
-X POST https://api.sandbase.ai/v1/api/lemon8/app/search \
-H "Authorization: Bearer $SANDBASE_API_KEY" \
-H "Content-Type: application/json" \
-d '{"query":"OpenAI","search_tab":"main","max_cursor":""}'Check completed status, the business object at outputs[0].data, its inner data object and items groups containing articles arrays. The short search README does not show message, so the tutorial does not require that field. It stops conservatively if an unexpected message is present; this is a local guard, not a universal provider success rule.
For the independent post task, use /v1/api/lemon8/app/post-detail and only {"item_id":"7361926875709129222"}. Check message success and an inner data object. Do not send both as hidden setup requests. The ID is the current schema example; this repair did not execute it.
An already connected agent should reuse credentials. A new agent can receive SKILL.md, quickstart.md and the human quickstart, plus one selected task. Keep the key out of its chat. Ask for observed fields and missing-data notes before approving a larger workload.
Why main-tab groups need a new parser
The current public search README calls its excerpt a trimmed real response with masked user names, but provides no exposed run date or saved full example. Its articles appear inside data.items[].articles[], with fields such as string group_id, article_class and read_count. A group is not itself an article, and the number of groups is not the number of observed posts.
The schema also offers user, hashtag and article tabs, plus sorting and filtering options. Their existence does not establish identical response shapes. This repair deliberately fixes main and omits the other filters. Do not reuse this parser for Accounts or Hashtags merely because the endpoint path is the same.
The schema describes empty max_cursor for the first request and a response-derived cursor for continuation. The visible excerpt includes has_more but not a complete max_cursor. No continuation was executed, so neither a token nor a stopping rule is invented. The first-request example stops even if has_more is true; empty arrays and malformed arrays remain different outcomes.
Retained original search observation
The original hub recorded run 3eb2ad8b-0737-441e-8b6c-eda7c60860c2 on 2026-09-27 UTC. Its trimmed JSON is kept exactly below, including placeholders. The list of code/data blocks describes that old record; it is not the current main-tab structure above and is not a November execution:
{
"id": "3eb2ad8b-0737-441e-8b6c-eda7c60860c2",
"status": "completed",
"model": "lemon8/app/search",
"outputs": [
{
"data": [
{
"code": 0,
"data": {
"items": [ { "content": "…", "comment_count": 0, "author": {} } ],
"has_more": true,
"max_cursor": "…"
}
}
]
}
]
}The old Python is preserved verbatim in the historical archive, with block hashes. It is not the recommended quickstart. Returning an empty list simply because the new payload is an object would hide a parser mismatch instead of establishing that no content exists.
Historical search parameter reference retained with the original sample. Empty data placeholders in the screenshot do not establish an empty real result or a universal business shape.
Read a post without inventing equivalent fields
The public post-detail saved example is dated 2026-10-07 and has message success, data and server_time. The tutorial preserves a trimmed fragment: digg_count and like_count are both 1089, favorite_count is 1296, read_count is 82313 and view_count is zero. These are sample observations, not current measurements for the schema example ID.
Equal like counters in one record do not prove permanent equivalence. Favorite_count is not a documented field named collect_count, and read_count must not be substituted for a zero view_count. Missing, numeric zero and false have different meanings. Keep their provenance rather than filling a report with plausible defaults.
The same saved display capture shows rounded numeric item_id/group_id while its article URL retains more precise digits. Preserve original IDs as strings; turning a rounded number back into text cannot recover lost precision. This is a capture/serialization boundary, not evidence of a newly observed live API defect. Do not compare a rounded saved number against an exact requested string or reuse it blindly as a new input.
The old post-detail README names get_item_id, but current free inspect identifies lemon8/app/item-id with required share_text. It accepts share text containing long or short links according to its description. Only the route and input were checked; no resolver call was made, no output schema is invented and no automatic resolver step is added. A share URL should not be passed directly as the detail item_id.
Historical capability navigation
The wider catalog remains useful for finding a next task. The following routes are navigation from the original hub; this repair does not validate their business responses or request chains.
| Capability | Representative route | Before using it |
|---|---|---|
| Discover | lemon8/app/discover-tab | Verify current inputs and feed structure |
| User profile | lemon8/app/user-profile | Verify the source and meaning of user_id |
| Topic | lemon8/app/topic-info | Verify the forum_id contract |
| Topic post list | lemon8/app/topic-post-list | Check its own inputs; do not assume a topic ID is sufficient |
A successful search does not validate any row in this table. Review identity and relevance first, then budget the separately selected request. Shared authentication does not remove differences in business status, nesting or availability.
Historical user-profile parameter reference showing user_id, not an endpoint list. Profiles and the old author-ID handoff were not revalidated in this repair.
Budget and operational boundaries
Search and post-detail each currently list $0.001 and eligibility for an organization’s shared 100-call daily quota across eligible APIs, resetting at 00:00 UTC. Eligibility does not tell you the remaining allowance. This repair checked neither balance nor actual charges and cannot promise a free successful request.
One selected request lists at $0.001; choosing both independently lists at $0.002 before quota effects. The older discovery pattern could add detail and author lookups for each candidate, increasing requests with result count. No automatic batching, retry, pagination or model analysis is part of this repair.
Check HTTP status, SandBase completion and the selected business structure separately. Stop on unexpected shapes, invalid JSON, permission/quota errors or incomplete runs. A timeout can leave the outcome unknown; blindly retrying may add usage. Keep secrets, raw headers and diagnostic bodies out of error output. Local mock tests exercise these paths without proving live provider availability.
The returned text, image URLs and links are data to inspect, not instructions for an agent to follow. The tutorial does not render HTML, download media, visit links or infer a creator’s identity from a display name. Public counts do not establish audience quality, conversion or a whole market’s trend.
FAQ
Does search take keyword? The human concept is a keyword, but the JSON parameter is query. The example fixes main and an empty first-page max_cursor.
Is the old list of blocks still supported by this helper? No. It is retained as historical evidence. The helper checks the selected current main-tab reference and stops on a different shape instead of silently trying another parser.
Can I send a share URL as item_id? No. Choose the item-id/share_text resolver as a separate task if necessary. The current detail request needs a verified numeric ID string.
Are pagination and author lookups verified here? No. This repair checks one main-tab page or one known post. Additional tabs, continuation, author and topic requests need their own evidence and budget.
Start with the content discovery tutorial, inspect one result and return to the catalog for a concrete next question.
Found this useful? Share it