Skip to content

Releases

v1.0.1

Choose a tag to compare

@github-actions github-actions released this 17 Nov 13:38

The GitHub organization FreshLabDev is now asterfield-dev, and the bot moves
with it. Nothing a person sees in Telegram changes.

Changed

  • Images are published as ghcr.io/asterfield-dev/searchy. The
    organization was renamed, and GitHub Container Registry does not redirect
    the old name: ghcr.io/freshlabdev/searchy answers 403. Point the
    deployment's image at the new path; every earlier digest is there under it
    too.
  • tg moves from v0.1.0 to v0.4.0-alpha.2, whose module path is
    github.com/asterfield-dev/tg. A press under one of searchy's inline
    results now arrives without a message instead of an empty one; the download,
    grid and menu buttons check for that before reading it, so such a press is
    answered as before.
  • Links, the module path and the docs point at github.com/asterfield-dev.
    Old links still redirect.

Known Limitations

  • Released as a stable patch without a pre-release first, at the owner's
    request: the change is the organization rename and the tg upgrade, checked
    by the full test suite, not by a run on a test bot.

Container image

ghcr.io/asterfield-dev/searchy@sha256:0f518eecf9ed7595b4da7a60b2a511ece70bbbb6259ecbd035f869e74eb671e6

v1.0.0

Choose a tag to compare

@github-actions github-actions released this 14 Nov 19:47

searchy searches by itself. The private SearXNG instance is no longer needed:
Bing, DuckDuckGo and Wikimedia are queried directly from Go, the first page
arrives while slower sources are still running, and inline scrolling goes
ten pages deep. This is v1.0.0-alpha.2, which has run the primary bot since
23 October, plus a Go security update. Coming from v0.2.3, read both alphas
under this section; the two things an operator has to act on are below.

Since 23 October alpha.2 answered 350 inline searches from 13 people, 18 in
DMs and 1 in a group, with no errors logged. The results people picked came
from Bing images (62), DuckDuckGo videos (30), Bing videos (25), DuckDuckGo
images (18) and Wikimedia (3).

Breaking

  • Operators. The default backend is native search, not SearXNG. Set
    SEARCH_BACKEND=searxng to keep the old one. See v1.0.0-alpha.1.
  • Operators. INLINE_CACHE_TIME defaults to 60 seconds. If .env sets
    it explicitly, lower it to 60 or remove the line. See v1.0.0-alpha.2.

Security

  • Go moves from 1.26.10 to 1.26.11: a crafted HTTP/1 request could crash
    the server through a race on the MaxBytesReader header (GO-2026-6859).
    This update went into the stable version without an alpha of its own; it
    changes no code of searchy's, and the tests and govulncheck pass on it.

Known Limitations

  • Native search has no YouTube, Dailymotion or Pinterest source. The SearXNG
    fallback adds them only when native results are weak.
  • DuckDuckGo Images answers with a JavaScript challenge often enough that
    image results come mostly from Bing; searchy pauses it for ten minutes
    each time and does not try to solve the challenge.
  • Wikimedia Commons answers HTTP 429 to fast bursts and recovers within
    minutes.

Container image

ghcr.io/freshlabdev/searchy@sha256:ee17e5c42eda40b94c0766663ae86a0a0832d792498c3ad61ea878385736bfac

v1.0.0-alpha.2

v1.0.0-alpha.2 Pre-release
Pre-release

Choose a tag to compare

@github-actions github-actions released this 23 Oct 19:05

Fixes for what alpha.1 got wrong on the primary instance: DuckDuckGo videos
were all dropped, inline scrolling stopped early, Telegram's cache was off for
most image searches, and a search could expire while it was being scrolled.

Changed

  • INLINE_CACHE_TIME now defaults to 60 seconds instead of 300, so a first
    page that Telegram serves from its own cache cannot outlive the search session
    behind its next_offset. If .env sets INLINE_CACHE_TIME explicitly,
    lower it to 60 or remove the line
    , otherwise the next scroll after five
    minutes can answer "search expired". This also applies to
    SEARCH_BACKEND=searxng.
  • The bot writes its name in lowercase in the menu, the about panel and the
    greeting in all 16 languages: "Hi! This is searchy".

Fixed

  • DuckDuckGo Videos returns results again. Its covers come from Bing's
    thumbnail service as /th/id/… links, which the JPEG check accepted only in
    the /th?id=… form, so every DuckDuckGo video was dropped and video search
    relied on Bing alone. Both now return about 50 videos per query.
  • When DuckDuckGo answers with a 202 challenge, that category is skipped for
    ten minutes and no longer marks the page degraded. DuckDuckGo Images gets
    the challenge on every request, so in alpha.1 almost every search that
    included images was sent with cache_time 1. In a 36-query searchbench
    run, degraded first pages went from 15 to 6.
  • Inline scrolling goes ten pages deep, 240 results at the default page size.
    alpha.1 stopped after about 50. MAX_RESULTS still sets the page size.
    CACHE_SIZE bounds all sessions together: at the default 2048, about eight
    searches scrolled to the end fit before the oldest idle ones are dropped.
  • A search session lives CACHE_TTL from the last scroll, not from the first
    query, so scrolling near the end of the window no longer returns "search
    expired". A new query never reuses a session older than CACHE_TTL.
  • A first page sent while a slower source is still running is no longer marked
    degraded. Only a source that failed sets cache_time to 1.
  • SEARCH_MAX_CONCURRENT now also limits simultaneous requests to sources.
    Values above 8 had no effect on them.

Removed

  • The f: prefix. It did exactly what an ordinary query does, and the help
    text in every language presented it as a separate mode. f: cats is now
    searched as typed.

Known Limitations

  • DuckDuckGo Images answers every request with an HTTP 202 JavaScript
    challenge, so image results come from Bing and Wikimedia only. The SearXNG
    duckduckgo images engine returned almost nothing in the October 22
    comparison runs as well. searchy does not try to solve the challenge.
  • Compared with the SearXNG backend, native search has no YouTube, Dailymotion
    or Pinterest source. SearXNG fallback adds them only when native results are
    weak.
  • Wikimedia Commons answers HTTP 429 to a fast burst of requests, such as a
    searchbench run, and recovers within minutes.
  • Inline paging, DM and group search and the full vido delivery matrix have
    still not been checked live in Telegram. CI and mocked tests do not cover
    those flows.

Container image

ghcr.io/freshlabdev/searchy@sha256:7f526692d48388b0c082fe911c9993042d474ca163cbdd52b6757df4edf246cf

v1.0.0-alpha.1

v1.0.0-alpha.1 Pre-release
Pre-release

Choose a tag to compare

@github-actions github-actions released this 22 Oct 22:14

Breaking

  • The default backend changes from private SearXNG to native Go metasearch.
    Set SEARCH_BACKEND=searxng to retain the previous backend. In native mode,
    CACHE_SIZE counts retained results, including frozen pages, rather than
    legacy cache entries.

Changed

  • Native Go metasearch is the default backend: direct Bing and DuckDuckGo
    images/videos, plus Wikimedia images. No AI model or separate Python search
    service is required. The existing SearXNG backend remains available with
    SEARCH_BACKEND=searxng for a compatible switchback.
  • Native inline retrieval collects a partial first page, freezes returned
    cards, and adds later results through stable opaque pagination. Telegram's
    already-answered inline picker is not updated in place. DM/group grids and
    owner-bound Vido delivery keep their existing behavior.
  • f: explicitly selects the same fast, non-AI search as an ordinary query.
    Combine it with i: or v: in either order, including in /search.
  • Native search defaults to an 800ms first-page collection window and 3.5s
    source budget, with up to eight concurrent searches. Optional SearXNG fallback
    stays within the overall 8s request budget; SEARCH_FALLBACK=none disables
    it. These are operating budgets, not measured production latency guarantees.
  • Result caches and pagination sessions are bounded in memory with a default
    five-minute TTL. Raw query text stays out of sessions, logs, and databases;
    partial/degraded responses are not cached as complete healthy results.
  • Local Compose retains private SearXNG for fallback and rollback without public
    ports. Native-only startup can run just the bot. Search needs no new database.

Known Limitations

  • This alpha is intended for owner-directed testing on the primary Searchy
    instance. Actual Telegram inline/media delivery and the full Vido delivery
    matrix remain to be verified after rollout; CI and mocked integration tests
    do not prove those live flows.
  • DuckDuckGo Images can return an HTTP 202 challenge. Other native sources can
    still produce partial results; the optional private SearXNG fallback covers
    weak or empty initial retrieval within the overall search budget.

Container image

ghcr.io/freshlabdev/searchy@sha256:8356b191b657e87b2aaee5104c9850a6bbc36c395ebc67f7600216237053ba80

v0.2.3

Choose a tag to compare

@github-actions github-actions released this 20 Oct 18:42

Searchy is built with Go 1.26.10, which closes nine standard-library advisories
its code reaches in the image running since September. Nothing changes in what
the bot does or how it is configured.

There was no pre-release: the bot's code did not change, only the Go release it
is compiled with and the release machinery below, and the patch Go releases
between 1.26.6 and 1.26.10 are security fixes.

Security

  • Go moves from 1.26.6 to 1.26.10 (go.mod and the build image). govulncheck
    found nine advisories reachable from Searchy on 1.26.6, among them
    GO-2026-6611, -6612, -6613 and -6617 in net/http and HTTP/2; on 1.26.10 it
    finds none.

Changed

  • A release waits for CI. release.yml runs ci.yml on the tagged commit
    and publishes nothing unless it passed. When CI cannot run, release.yml can
    be started by hand with the tag and a reason; that run skips CI and prints the
    reason at the top of the release notes. See docs/releases.md.
  • The release notes end with the image digest to pin, which docs/releases.md
    already said they did.
  • The image carries an org.opencontainers.image.revision label beside the
    version, so docker inspect names the build without starting it.
  • CI also runs weekly, so a new advisory against a dependency or the standard
    library is noticed while nothing is being pushed.
  • Dependabot opens grouped version updates against dev.

Container image

ghcr.io/freshlabdev/searchy@sha256:9d0b47e7d6367fceb159ae081e21d2fcace3876d558e04bef218f669b9bc2437

v0.2.2

Choose a tag to compare

@github-actions github-actions released this 10 Sep 08:49

A language picked by hand can be handed back, and the sixteen-button grid stops
looking broken.

Only the selected language carried a glyph, so fifteen labels started at the
left edge and one started two characters in. Close is now written in one place
and painted destructive, and the three buttons that carry meaning are the only
coloured ones.

Changed

  • docs/releases.md names the real cause of the 403 a digest pull used to
    answer: the account running the deploy could not read the private package,
    not anything about the digest. Pulling the tag first was a workaround for
    that, and is no longer required.

v0.2.2-alpha.1

v0.2.2-alpha.1 Pre-release
Pre-release

Choose a tag to compare

@github-actions github-actions released this 09 Sep 20:15

Added

  • "Follow Telegram" on the language screen. Picking a language wrote a
    manual claim to the shared core store, and a manual claim outranks
    everything β€” but nothing removed it, so one mistaken tap followed a person
    across every bot in the family until an admin deleted the row by hand. The
    new button under the grid deletes searchy's own observation and lets the
    Telegram client's language_code win again. Translated into all 16
    languages.

Changed

  • Every option of a set is marked, in both states. The language picker
    marked only the current language, so fifteen labels started at the button's
    left edge and one started two characters in. Both states now carry a glyph
    (β—‰ / β—Ž ), the way the statistics tabs already did.
  • Three buttons carry colour, and no others. Close is destructive
    everywhere it appears β€” including the result grid, which had its own copy of
    it. The current language and the open statistics tab are marked as state.
    Language on the DM /start panel is the one thing that leads, because
    nothing else on that screen is legible until the language is right. The group
    panel leads with inline search by position and takes no colour at all. Older
    Telegram clients ignore button styles and show the panels exactly as before.
  • Every screen's text comes from the same two helpers. The About card, the
    result grid caption and the "Bot" sub-heading in global statistics each wrote
    their own bold/italic markup; the empty statistics screen put its one line
    outside the blockquote, so the panel changed shape depending on whether you
    had searched yet. Only that empty screen looks different β€” it gains the
    blockquote every other panel has.

v0.2.1

Choose a tag to compare

@github-actions github-actions released this 09 Sep 16:09

Searchy leaves the third-party Telegram client it was the last bot in the family
to use, and moves onto the shared one. Nothing about searching changes; what
changes is that it can now be given the things the shared client gets.

Changed

  • Depends on the stable github.com/FreshLabDev/tg v0.1.0.

  • /start is a different screen in a group than in a DM. A DM is the
    user's own space, so it keeps language and personal statistics and has no
    Close β€” there is nothing else in the chat, and the button offered to delete
    the only thing on screen. A group is shared, so the panel there leads with
    inline search, keeps Help and About, points at the DM for the personal
    settings, and always offers Close: the panel is one member's menu sitting in
    everyone else's feed. Help, About, Language and Statistics follow the same
    rule β€” Close only where there is something to close.

  • The About tab is the family's format. Name and version on one line, one
    line of purpose, then a blockquote of key Β· value facts: the search backend,
    the build date, the repository as a link with its license, and the admin. The
    repository is a link inside the text rather than a button of its own, because
    two controls for one action is one too many. Everything is translated into all
    16 languages, values included.

  • The build date is visible. aboutBody had been passing buildinfo.Date into
    a string that referenced it in none of the 16 languages, so nothing told a CI
    build from a laptop one.

  • Emoji are gone from every button and heading, in all 16 languages: About,
    Help, Language, Statistics, Search, Download, Video download settings, Open
    original, Open on {platform}, Retry sending, the /start greeting, the panel
    headings and the results caption. They stay only where they mark state β€” the
    selected language and the active statistics tab keep β—‰ / β—Ž. An icon on every
    control is decoration that stops meaning anything, which is why the marks that
    do mean something got lost in it.

  • Navigation reads Back and Close, without the β¬… and βœ–.

  • homePanel is assembled the way every other panel is, through header() and
    blockquote(), instead of concatenating three keys by hand.

  • Searchy runs entirely on github.com/FreshLabDev/tg.
    github.com/go-telegram/bot is gone: the poll loop, every send, the inline
    answers and the Vido delivery bridge all go through the family's own client.
    Searchy was the last bot on a third-party library, which cost it everything
    the shared client had learned β€” Bot API 10.3 button styles, ephemeral
    messages, one place for token redaction, one retry policy, one set of error
    classifiers β€” and the gap was only going to widen.

  • The dispatcher is now searchy's own long-polling loop. Updates are still
    handled concurrently (WORKERS, default 32), because the inline path
    deliberately blocks: the debouncer holds a keystroke for its window and
    abandons it when a newer one arrives, which only works while both are in
    flight. The offset advances as soon as an update is dispatched, as it did
    before, and a failed poll backs off from one second to thirty rather than
    ending the process.

  • Preflight now lists everything searchy cannot work without, not just two
    methods: answerInlineQuery, answerCallbackQuery, deleteMessage,
    editMessageMedia, editMessageReplyMarkup, editMessageText,
    sendMessage and sendPhoto β€” plus sendVideo, sendAudio, sendDocument
    and sendMediaGroup when the Vido bridge is on, since those are only
    reachable through it. A Bot API server too old for any of them now refuses
    the start instead of letting searchy poll happily and answer nothing, which
    is the failure the check exists for.

  • A Vido plan's local_file_uri is sent as a local path rather than a bare
    file:// string, so a searchy pointed at Telegram's own endpoint is told
    that the path could never work there, instead of getting back a
    wrong-file-identifier error that points nowhere near the cause.

  • Delivery failures are classified from Telegram's HTTP status rather than a
    library's sentinel errors. The definite failure set is unchanged (400, 401,
    403, 404); everything else still becomes delivery_unknown, so a send that
    may have landed is never silently retried.

  • Captions and text in a delivery plan are always sent as HTML. The plan field
    is named caption_html and its text is Vido's own, so the previous
    parse_mode passthrough only ever chose between HTML and rendering HTML
    literally.

  • One versioning and release document for the whole family. docs/versioning.md
    and docs/releases.md are now byte-identical across every Asterfield
    repository apart from two clearly marked sections: this repository's own
    version line, and the surface where a change here breaks something. They spell
    out what each of the three numbers means, what the -alpha.N suffix counts,
    when alpha becomes beta and when it is legitimate to skip to rc or run a
    pre-release in production.

  • Pre-releases are now tagged on dev, not main. Only stable versions are
    tagged on main, on the merge commit from dev. release.yml had no branch
    check at all before, so a tag pushed from any branch would publish; it now
    refuses a tag that is not on the branch its channel is published from.
    Earlier pre-releases were tagged on main under the previous rule; they are
    left as they are.

  • The production stack pulls the released image instead of building one.
    deploy/ws04/searchy/compose.yaml built from a working copy on the host, so
    what served users was not the artifact CI had tested, scanned and published,
    and the version it reported came from SEARCHY_VERSION / SEARCHY_COMMIT /
    SEARCHY_BUILD_DATE kept by hand in .env. Those are now baked into the
    image by the release workflow, and SEARCHY_IMAGE -- which has no default,
    so an unset one stops the stack -- names the GHCR reference to run.

v0.2.0-alpha.4

v0.2.0-alpha.4 Pre-release
Pre-release

Choose a tag to compare

@github-actions github-actions released this 08 Sep 12:37

Changed

  • Startup goes through github.com/FreshLabDev/tg, the client shared by the
    bot family. Searchy keeps go-telegram/bot for polling and handlers, but
    the 170 lines of hand-rolled HTTP around it -- getMe with its own retry
    ladder, deleteWebhook, endpoint building and two error types -- are gone,
    along with the knowledge that had been living only in this repository.
  • go-telegram/bot moves from v1.21.0 to v1.25.0, which is the release that
    covers Bot API 10.3.
  • TELEGRAM_BOT_API_BASE_URL is validated at startup. It used to be checked
    while building every request; the shared client takes it as given, so a typo
    now fails with a message naming the variable instead of a connection error on
    the first poll.

Added

  • A startup preflight naming answerInlineQuery and sendPhoto. Searchy is
    an inline bot that answers with pictures, and a Bot API server without those
    would leave it polling and never answering.

Security

  • golang.org/x/image moves to v0.45.0. The advisory against v0.43.0 is
    reachable straight from collage.Render through the WebP decoder, and every
    image searchy decodes comes from a search result -- that is, from the
    internet.
  • The Go floor moves to 1.26.6, build image included.

v0.2.0-alpha.3

v0.2.0-alpha.3 Pre-release
Pre-release

Choose a tag to compare

@github-actions github-actions released this 09 Aug 07:33

Fixed

  • Retry the final Vido bridge commit after Telegram operations were confirmed,
    preventing transient Core errors from leaving delivered media leased.

Operations

  • Requires Core v0.2.0-alpha.1 migration 008 and Vido
    v2.3.7-alpha.5 before bridge intake resumes.