# Tweet Archive

> Paste a tweet URL to view it and get a permanent Wayback Machine snapshot, so
> the tweet survives even if it's deleted from X. Free, no account required.

How to use: 1) Paste a tweet URL on the homepage 2) View the tweet 3) A
Wayback Machine capture is automatically requested in the background.
Pricing: Free, unlimited, no account required.

## How it works
- Reads tweets via X's free public syndication API. Regular tweets show in
  full; X Premium "long tweets" are truncated to a preview, the same
  limitation X's own oEmbed widget has. Callers with their own X API
  access can avoid this -- see x_bearer_token below.
- Every view of a tweet's page schedules an archive.org (Wayback Machine)
  capture in the background, throttled so repeat views don't spam
  archive.org.
- A tweet can have many archived versions, not just one: it's resubmitted
  whenever its text or media changes, and every past capture stays listed
  on the tweet's page.
- If a tweet is later found removed from X, its content stops being shown
  and only the link to its most recent archived snapshot remains.
- A daily background check revisits tweets that stop getting organic views,
  so archives stay current even without traffic.

## API

AI agents and scripts can archive tweets directly, no key or account
required.

`POST /api/v1/tweets`
Body: `{"urls": ["<tweet URL or bare numeric ID>", ...]}`, 1-10 items per
request. Returns a JSON array, same order as the input, one entry per URL:
either the tweet (see shape below) or `{"input": "...", "error": "..."}`
for anything unparseable or unfetchable. A bad item never fails the rest
of the batch. Archiving happens in the background after the response --
snapshots may still be empty right away; poll the GET endpoint below.

Without a token, text over ~280 characters gets trimmed (see "How it
works" above). If you have your own X API access, pass it: it costs
nothing extra for a normal-length tweet, and it means the tweet gets
archived in full instead of truncated -- a better permanent record for
anyone who looks it up later, not just for you.

Two optional fields apply to the whole request:
- `x_bearer_token`: your own X API Bearer Token, for full-length tweets.
  Only spent when actually needed -- the free API runs first, and your
  token is only used if that result is truncated or fails outright, so a
  normal-length tweet costs you nothing even if you supply one.
- `agent`: a name to self-identify as (e.g. "my-bot/1.0"), recorded on
  the archive for attribution. Falls back to your User-Agent if omitted.

`GET /api/v1/tweets/{id}`
Read-only status/poll endpoint, no side effects. 404 if the ID has never
been seen.

`POST /api/v1/tweets/{id}/thread/check` and `.../thread/ingest`
Pull in the tweets around `{id}`, not just that one tweet: what it's
replying to (ancestors -- always free, no key, walked backward one tweet
at a time) and/or the author's own continuation tweets (forward -- needs
`x_bearer_token`; `check` pays for exactly one X API search call to find
the author's latest reply in the conversation, then walks backward from
it for free, so a whole thread costs about one search call regardless of
length). `check` returns a free preview (every discovered tweet's text,
whether it's truncated, and a real cost estimate for upgrading the
truncated ones); `ingest` actually archives everything -- calling it at
all is the confirmation, no separate confirm flag. Full request/response
shapes are in /SKILL.md.

Response shape (one tweet):
```
{
  "id": "2095401070080184644",
  "status": "live" | "removed",
  "tweet": { ...raw syndication payload... } | null,
  "truncated": false,
  "note": "present only when truncated -- explains how to get the full text",
  "archive_url": "https://x-dot-com-archive.vercel.app/tweets/2095401070080184644",
  "first_archived_at": "2026-09-01T09:00:00.000Z",
  "removed_at": null,
  "removed_snapshot": null,
  "snapshots": [
    { "wayback_url": "...", "archived_at": "...", "screenshot_url": "..." | null, "event": "created" | "edited" }
  ]
}
```
`snapshots` is newest-first; `snapshots[0]` is the latest capture,
`snapshots.length` is the count. `tweet` is `null` once a tweet has been
removed from X. `first_archived_at` is when we first archived it (not the
tweet's own X creation time). `removed_at`/`removed_snapshot` are set once
removed -- `removed_snapshot` is the archived version live right before
that; both null while live, and `removed_at` can be null even when
`status` is `"removed"` for tweets removed before this field existed.

No authentication and no rate limiting yet -- be a reasonable citizen.
Rate limiting is planned; this section will be updated when it ships.

A ready-to-use skill file for agent frameworks: /SKILL.md
An OpenAPI 3.1 spec (for GPT Actions and similar tool-calling frameworks): /openapi.json
Human/agent-readable docs with examples: /agents