Amended Endpoints — what the feed release changes in endpoints you already have
Base URL: https://dev.api.nexgate.co (staging) · https://api.nexgate.co (production)
Short Description: This document lists only the endpoints that already existed and have now changed. Nothing here is a new URL. Every path below is one your app already calls today, and every entry says exactly what moved under it, whether the change is visible in the response, and whether you have to do anything about it.
Its sibling document is
09_NEW_ENDPOINTS.md— the endpoints and controllers that did not exist before (home feed, reels, marketplace, events, interactions, and the admin/ops APIs). If a path is not in this file, it was not amended; if it is not in that file, it is not new.
Hints:
- No path, verb or field was removed. Every change is additive, a bug fix, or a stricter parameter.
- Two response fields were added (both on
ClipResponse). An app that ignores them behaves exactly as it does today. - Three endpoints return fewer items than before for some viewers. That is the point: they were leaking private content.
- Four old paths are being deleted, not migrated from — see Old endpoints being deleted. There is no production data and no grace period: when the frontend has switched, they go.
- Search this file by path. Every heading is the literal URL.
How to read this document
Each entry carries two chips.
Change type — what kind of amendment it is:
| Chip | Meaning |
|---|---|
| SECURITY FIX | The endpoint used to return content the viewer was not allowed to see |
| RESPONSE | A field was added, or a field that used to be empty is now filled |
| BEHAVIOUR | Same shape, different contents, ordering or side effect |
| PARAMS | A query parameter is now validated or clamped |
| BEING DELETED | Still answers today, but is scheduled for removal — do not build on it |
App action — what the frontend must do:
| Chip | Meaning |
|---|---|
| 🟢 None | Ships safely with no app change |
| 🟡 Optional | Works untouched, but there is something worth picking up |
| 🔴 Required | The app must handle this or it will misbehave |
Index
| # | Endpoint | Change type | App action |
|---|---|---|---|
| 1 | GET /api/v1/e-social/feed |
RESPONSE · BEING DELETED | 🔴 Required |
| 2 | GET /api/v1/e-social/clip/feed |
SECURITY FIX · RESPONSE · BEHAVIOUR | 🔴 Required |
| 3 | GET /api/v1/e-social/clip/{mediaId} |
SECURITY FIX · RESPONSE | 🟡 Optional |
| 4 | GET /api/v1/e-social/clip/user/{userId}/videos |
SECURITY FIX · RESPONSE · BEHAVIOUR | 🔴 Required |
| 5 | POST /api/v1/e-social/clip/{mediaId}/view |
BEHAVIOUR | 🟡 Optional |
| 6 | GET /api/v1/e-social/hashtags/{hashtag}/posts |
SECURITY FIX · PARAMS · BEHAVIOUR · RESPONSE | 🟢 None |
| 7 | GET /api/v1/e-social/hashtags/trending |
BEHAVIOUR · PARAMS | 🟢 None |
| 8 | GET /api/v1/e-social/hashtags/search |
BEHAVIOUR · PARAMS | 🟢 None |
Standard Response Format
Unchanged. Every endpoint below still answers in the usual envelope:
{
"success": true,
"httpStatus": "OK",
"message": "Operation completed successfully",
"action_time": "2026-09-18T10:30:45",
"data": { }
}
Errors are unchanged too: 400 bad request, 401 token problems, 403 permission, 404 not found, 422 validation, 500 server. Where an amendment introduces a new reason for an existing status, the entry says so.
Endpoints
1. Home feed (old)
Endpoint: GET /api/v1/e-social/feed
Change type: RESPONSE BEING DELETED · App action: 🔴 Required — move to /api/v1/feed/home
This endpoint is going away. It is listed here because it still answers today and its contents changed. Do not build anything new on it.
Access Level: 🔒 Protected · Authentication: Authorization: Bearer <access token>
What changed
a) topComments is now always populated.
This endpoint used to call the post mapper with "no top comments", so topComments came back empty on every post. The mapper now always fills it (up to 5, most-liked first) because the feed needs it and loading it costs nothing extra in the batched mapper.
b) A backend switch can make it answer from the new pipeline. There is a feed.legacy.fromPipeline setting that makes this endpoint serve the new feed in the old shape. It was built for a migration that is no longer needed, and it is not part of the plan — it will be deleted together with the endpoint. Mentioned only so nobody is surprised to find it in the code.
Response shape (unchanged)
{
"success": true,
"httpStatus": "OK",
"message": "Feed retrieved",
"action_time": "2026-09-18T10:30:45",
"data": {
"posts": [ { "postId": "…", "topComments": [ { } ] } ],
"unseenCount": 12,
"nextCursor": "eyJ0IjoxNzU4…"
}
}
What the app must do
Move to GET /api/v1/feed/home. See 09_NEW_ENDPOINTS.md A1 and 07_FEED_ENDPOINTS_GUIDE.md. This path is deleted once you have.
Until then it keeps working as it does today.
2. Clip feed
Endpoint: GET /api/v1/e-social/clip/feed
Change type: SECURITY FIX RESPONSE BEHAVIOUR · App action: 🔴 Required
Access Level: 🔒 Protected · Authentication: Authorization: Bearer <access token>
What changed
a) It no longer leaks private videos. The feed used to return clips from followers-only and mentions-only posts to anyone, and clips from accounts the viewer had blocked. It now filters by the post's visibility, the viewer's follow relationship, mentions, and blocks — in the database, before paging.
b) Moderation labels apply. A clip whose post or author carries a moderation label is either dropped from the page or marked for an interstitial, per the rules in 01_ARCHITECTURE.md §9.2. Followers are treated more leniently than strangers.
c) Two new fields in each clip (see the table below).
d) A page can be shorter than limit while hasMore is still true. The cursor is taken from the last row the database returned, before labels remove anything. So clips.length < limit no longer means "end of feed" — only hasMore === false and nextCursor === null mean that.
e) limit is clamped to 1–20 (it was previously used as given).
Query Parameters
| Parameter | Type | Required | Description | Validation | Default |
|---|---|---|---|---|---|
cursor |
string | No | Opaque page cursor from the previous response | Invalid cursor → 400 |
none (first page) |
limit |
integer | No | Clips per page | Clamped to 1–20 (new) | 10 |
New response fields
| Field | Type | Values | Description |
|---|---|---|---|
visibility |
string | ALLOW, INTERSTITIAL |
INTERSTITIAL means: do not autoplay. Show a "sensitive content — tap to view" cover first. Defaults to ALLOW |
media.durationMs |
integer | milliseconds | Clip length. Useful for the progress bar, and needed to report watch time correctly in the interactions batch |
Success Response JSON Sample
{
"success": true,
"httpStatus": "OK",
"message": "Clip feed fetched",
"action_time": "2026-09-18T10:30:45",
"data": {
"clips": [
{
"postId": "4f2a77e1-0c8b-4d2f-9a11-7b6c3e5d8f90",
"visibility": "ALLOW",
"media": {
"id": "9b1c2d3e-4f5a-6b7c-8d9e-0f1a2b3c4d5e",
"durationMs": 14200,
"variants": { "hls": "https://…/master.m3u8", "thumb": "https://…/thumb.jpg" }
},
"engagement": { },
"userInteraction": { }
}
],
"nextCursor": "MjAyNi0wOS0xN1QxMDoyMjozMTo0ZjJh…",
"hasMore": true
}
}
What the app must do
- Stop using
clips.length < limitas the end-of-feed test. UsehasMore. - Honour
visibility.INTERSTITIALclips must not autoplay behind no cover. - Expect fewer clips than before for some viewers. That is correct, not a regression.
3. Single clip
Endpoint: GET /api/v1/e-social/clip/{mediaId}
Change type: SECURITY FIX RESPONSE · App action: 🟡 Optional
Access Level: 🔒 Protected · Authentication: Authorization: Bearer <access token>
What changed
- A clip the viewer may not see now returns
404, with the messageClip not found. It used to be returned in full to anyone holding the id. This covers followers-only and mentions-only posts, blocked authors, and clips dropped by a moderation label. - Same two new fields as the feed:
visibilityandmedia.durationMs.
Error Response JSON Sample
{
"success": false,
"httpStatus": "NOT_FOUND",
"message": "Clip not found",
"action_time": "2026-09-18T10:30:45",
"data": "Clip not found"
}
What the app should do
Treat 404 here as "this clip is gone or not for you" and return the user to the previous screen — it is no longer safe to assume a 404 means deleted.
4. A user's videos
Endpoint: GET /api/v1/e-social/clip/user/{userId}/videos
Change type: SECURITY FIX RESPONSE BEHAVIOUR · App action: 🔴 Required
Access Level: 🔒 Protected · Authentication: Authorization: Bearer <access token>
What changed
Exactly the same three things as the clip feed (entry 2), on a profile's video grid:
- private and blocked-author videos are filtered out; a viewer looking at someone else's profile now sees only what they are entitled to (the owner still sees their own);
visibilityandmedia.durationMsadded;- short pages with
hasMore: trueare possible, andlimitis clamped to 1–20 (default 12).
What the app must do
Use hasMore, not the page length, to decide whether to load more in the profile grid.
5. Record a clip view
Endpoint: POST /api/v1/e-social/clip/{mediaId}/view
Change type: BEHAVIOUR · App action: 🟡 Optional
Access Level: 🔒 Protected · Authentication: Authorization: Bearer <access token>
Response: still 204 No Content, still no body. The request contract has not changed.
What changed
- A view is now counted once per viewer, cheaply, in Redis. Repeat calls from the same viewer are accepted and ignored instead of inflating the count.
- The
clip_viewstable is no longer written. Per-view rows are gone; the counter is what survives. - The request body is accepted and ignored. Watch time does not belong here — it belongs in the interactions batch, where the recommendation model can read it (
05_INTERACTIONS_CLIENT_GUIDE.md).
What the app should do
Keep calling this for the view counter, and send watch time as a VIEW event with watchMs and mediaDurationMs to POST /api/v1/feed/interactions/batch. Anything sent in this body is thrown away.
6. Posts for a hashtag
Endpoint: GET /api/v1/e-social/hashtags/{hashtag}/posts
Change type: SECURITY FIX PARAMS BEHAVIOUR RESPONSE · App action: 🟢 None
Access Level: 🌐 Public (a signed-in viewer sees more) · Authentication: optional bearer token
What changed
a) It no longer leaks private posts. The old version loaded every post id carrying the tag into memory and checked only "published and not deleted" — so followers-only and mentions-only posts reached anyone who opened the tag page. Visibility, mentions and blocks are now enforced in the database query itself.
b) Paging is real. The whole tag is no longer loaded into memory; the page and its total count are computed in the database.
c) Ordering is by published_at descending (it was createdAt). For a scheduled post these differ: the post now sorts by when it went live, which is what a reader expects.
d) page and size are clamped instead of being trusted.
e) topComments is now populated on each post, for the same reason as entry 1.
Query Parameters
| Parameter | Type | Required | Description | Validation | Default |
|---|---|---|---|---|---|
filter |
enum | No | ALL, PRODUCTS, SHOPS, EVENTS |
unknown value → 400 |
ALL |
page |
integer | No | 1-based page number | Below 1 is treated as 1 (new) | 1 |
size |
integer | No | Posts per page | Clamped to 1–50 (new) | 20 |
What the app should do
Nothing. Expect a signed-out viewer to see fewer posts than a signed-in one, and a blocked author's posts never to appear.
7. Trending hashtags
Endpoint: GET /api/v1/e-social/hashtags/trending
Change type: BEHAVIOUR PARAMS · App action: 🟢 None
Access Level: 🌐 Public · Authentication: none
Response shape: unchanged — a list of { hashtag, posts }.
What changed
a) Trending is now hard to game. Tags are ranked by how many different accounts used them, then by how many posts. One account posting the same tag fifty times counts once, and #sale #sale inside one post counts once.
b) Private and draft posts never count. Only public, published, non-deleted posts inside the window — so a private post can no longer push its tag onto a public page.
c) Results are cached for 5 minutes per window. The page is public and read far more often than it changes.
Query Parameters
| Parameter | Type | Required | Description | Validation | Default |
|---|---|---|---|---|---|
window |
string | No | 24h, 7d, 30d. Anything else is treated as 24h |
— | 24h |
limit |
integer | No | How many tags | Clamped to 1–50 (new) | 15 |
8. Hashtag suggestions (type-ahead)
Endpoint: GET /api/v1/e-social/hashtags/search
Change type: BEHAVIOUR PARAMS · App action: 🟢 None
Access Level: 🌐 Public · Authentication: none
Response shape: unchanged — a list of { hashtag, posts }.
What changed
- Answered from a Redis suggestion index, not from the database. The old query scanned and grouped the entire
post_hashtagstable on every keystroke. - Suggestions come from public posts only. Tags that exist only on drafts or private posts are no longer suggested.
limitis clamped to 1–20 (default 5).- An empty or blank
qreturns an empty list instead of an error.
Query Parameters
| Parameter | Type | Required | Description | Validation | Default |
|---|---|---|---|---|---|
q |
string | Yes | What the user has typed so far | Blank → empty list | — |
limit |
integer | No | How many suggestions | Clamped to 1–20 (new) | 5 |
Old endpoints being deleted
There is no migration and no grace period. This is a staging platform with no production users and no data worth keeping — the database will be dropped. So these paths are not being "migrated away from" over time; they are deleted from the codebase in one commit, as soon as the frontend confirms it has switched.
What that means for you:
- There is no fallback. Once the commit lands, these paths return
404. - There is no dual-running period to lean on, and no backend flag that will quietly keep them alive.
- Tell the backend team when your calls have moved. The deletion happens on that word.
The strings below are literal — copy them into your project search to find every call site.
| Search for this | What it is | Replace with | Notes |
|---|---|---|---|
api/v1/e-social/feed |
Old home feed | api/v1/feed/home |
Response shapes differ — see 09 A1 |
api/v1/e-social/feed/stream |
Feed SSE stream | nothing — use api/v1/events |
The chat/events stream now carries a feed.new hint; on receiving it, call api/v1/feed/home/has-new |
api/v1/e-social/feed/heartbeat |
SSE keep-alive | nothing | Dies with the stream above |
api/v1/e-social/clip/feed |
Clip feed | api/v1/feed/reels |
Still amended in the meantime (entry 2), so it behaves correctly right up to the day it goes |
Not going anywhere — keep calling these: api/v1/e-social/clip/{mediaId}, api/v1/e-social/clip/{mediaId}/view, api/v1/e-social/clip/user/{userId}/videos, and all three api/v1/e-social/hashtags/… endpoints.
A one-liner for a frontend repo:
grep -rn "e-social/feed\|e-social/clip/feed" src/ --include=*.{ts,tsx,js,jsx,dart,kt,swift}
Every hit is a call site that must move. When that grep returns nothing, tell the backend team — that is the signal to delete.
What did not change
Worth stating plainly, so nobody goes looking:
- No existing response field was renamed or removed, anywhere.
PostResponseis unchanged. Post, profile, search, bookmark and comment endpoints return exactly what they returned before. They are faster (a page of posts went from about 242 database queries to about 24) and they load the viewer's account once instead of twice, but the JSON is the same.- Engagement numbers (likes, views, comments) keep the same field names. Some values were recounted and corrected as part of an earlier release.
- No new notification is produced by any endpoint in this document.
- Authentication, the response envelope and every error code behave as before.
| Doc | What it covers |
|---|---|
09_NEW_ENDPOINTS.md |
Every new endpoint and new controller |
07_FEED_ENDPOINTS_GUIDE.md |
The app team's walkthrough of the new feed screens |
05_INTERACTIONS_CLIENT_GUIDE.md |
What the app must report back, and exact action definitions |
01_ARCHITECTURE.md |
The design. §A.3 is the migration and the deletion list; §9.2 is the label rule |