- New page "Manage LangBot with lbctl" (zh/en/ja): what lbctl replaces,
status/doctor, start/stop/logs, upgrade semantics, adopting an existing
Docker Compose deployment, autostart, install options, multiple
instances, uninstall and optional resource management.
- Update guide, Docker guide, troubleshooting and the one-line install
page now point to lbctl first and keep the manual steps as fallback.
- Remove the official plugin source index note from the README Quick
Start.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs: recommend the one-line lbctl installer instead of uvx
Every README now leads self-hosting with the one-line installer
(install.sh / install.ps1), which installs lbctl and deploys LangBot
natively or with Docker. The uvx one-liner is removed from the READMEs and
kept only as a development path in dev-docs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs: make the one-line lbctl installer the recommended deployment
- Add a "One-Line Install (Recommended)" page in zh/en/ja and list it first
under LangBot deployment.
- Turn the uvx package-manager page into a development-only path with a
warning pointing to the installer.
- Lead the update guide with `lbctl upgrade`; point the getting-started
guide and feature list to the installer; replace uvx quick starts in
articles.
- Mention `lbctl install --extra seekdb` next to the SeekDB notes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(recipes): point the self-host prerequisite to the one-line installer
Regenerated all 420 recipe pages from scripts/generate-recipes.mjs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(tests): count the 3 new one-line install pages in static build checks
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* ci: streamline pull request checks and run the full matrix daily
Refs #3055
* ci: run runtime contracts only on affected Core checks
* test(ci): distinguish SDK checks from SDK publication
* fix(ci): retain legacy regressions and scope production image builds
* docs(ci): clarify focused regressions and production image selection
* test(artifact): cover mixed media, isolation and native retention
Exercise real Core, Runtime and Runner with mixed attachments and retention. Keep deterministic checks separate from native 30-hour soak and credentialed provider acceptance.
* test(artifact): isolate compaction evidence in shared SQLite suite
* test(artifact): exercise native media compaction and scheduled GC under load
* test(artifact): allow native GC scheduler grace at soak finish
Run affected persistence/network, permission lifecycle and production-image upgrade/replacement acceptance before the existing required Core smoke status. Share persistence acceptance with Daily Endurance and fail the required status for missing, unsuccessful or cancelled P0 results.
Add authorization and content boundary tests and record a bounded mutation evaluation: 102 selected variants, with 87 killed and 15 explained survivors, stable across two executions. Product implementation and application dependencies are unchanged.
lbp build packed a developer's local venv/ into the .lbpkg. AutoLearn 0.1.1
shipped an 81 MiB package this way, past Core's 64 MiB marketplace download
limit, so every install failed with "Remote response exceeds the
67108864-byte limit". The runtime builds its own .venv from
requirements.txt, so a bundled venv is never needed.
Exclude venv/, .venv/ and any directory containing pyvenv.cfg.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
The lazy-route Suspense fallback was a bare `<div>Loading...</div>`, which
rendered as unstyled text in the top-left corner while the /home and
/wizard chunks loaded. HomeLayout then rendered an empty div during the
Workspace bootstrap.
- Add PageLoading (centered LangBot mark + spinner, theme-aware, fades in
after 200ms so fast loads show nothing) and ContentLoading.
- Use PageLoading for route fallbacks and HomeLayout's bootstrap phase so
the two phases look like one continuous state.
- Wrap the /home content area in its own Suspense so the sidebar/header
stay visible while a page chunk loads.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
Startup used pip.main to install missing packages into whatever environment
Core was running in, then exited asking for a restart. Every supported install
(Docker image, uv venv, uvx, lbctl-managed native) ships a locked dependency
set, so this only mutated ephemeral or foreign environments.
- check_deps stays as a diagnostic; missing packages now print reinstall
guidance and exit 1 instead of being installed.
- Remove the unused precheck_plugin_deps (legacy plugins/ requirements) and
utils/pkgmgr.py, which only served it and hardcoded the Tsinghua index.
- Remove the unmaintained docker/deploy-k8s-test.sh helper.
The pip runtime dependency stays: the plugin runtime installs plugin
requirements with `python -m pip`.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
Show a one-time milestone card once the current Workspace has handled
100 real executions (agent runs + pipeline queries). The answer is kept
per browser; 'Maybe later' snoozes for 7 days, Star/close are final.
The executions endpoint gains summary=false so the count-only probe
skips the dashboard aggregates.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
The pipeline launcher id (and anything persisted from it, such as the
HumanTakeover plugin's session target) is the customer's external_userid
only, while WeComCS send_message required "external_userid|open_kfid".
Manual replies therefore failed with
"ValueError: WeComCS target_id must be formatted as ...".
send_message now accepts a bare external_userid: it uses the open_kfid the
customer last wrote to, and otherwise falls back to the app's only customer
service account (kf/account/list). Ambiguous or malformed targets still
raise a clear error. Launcher ids are unchanged, so existing sessions,
access-control lists and routing rules keep working.
Fixes#3021
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* feat(web): show release notes from langbot.app changelog in the update dialog
Read the curated changelog feed at https://langbot.app/changelog.json for
update detection and release notes, and redesign the dialog as a timeline
of every release between the running and latest version, grouped by
change type. Clicking the sidebar version opens the current version's
notes. Falls back to the Space-cached GitHub releases when the feed is
unreachable.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* style(web): simplify update dialog to match the existing UI
Drop the header gradient, icon tile, coloured type pills and timeline
rail; use plain DialogHeader, neutral type labels and separators.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* style(web): use langbot.app changelog type tags in the update dialog
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* fix(web): order pre-release, post and hand-edited Core versions correctly in the update prompt
The sidebar badge and the dialog used two different comparators that
both read PEP 440 suffixes like 4.11.8b1 as 4.11.8, so beta users were
never offered the matching stable release, and the two could disagree.
Unparseable versions such as "custom" were offered every release.
Use one parser (coreVersion.ts) for both: stable X.Y.Z supersedes its
own a/b/rc/dev and -beta.N builds, .postN and +local map to their base
release, and unparseable versions never get an update prompt.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>