Some bugs fixed, new onboarding process for self-hosters and a new "speed config" dial to change the model mix are the lighthouse changes here.
What's Changed
- fix(e2e): assert terminal DOM for four recorded CI flakes by @FExB17 in #1592
- Fix/webfetch and misc by @orgaralf in #1593
- chore(deps): update ollama/ollama:latest docker digest to 075246f by @renovate[bot] in #1581
- test(e2e): wait for Vite after restart in make test-e2e by @FExB17 in #1580
- feat(summary): live-model eval harness for the rolling conversation summary (Sprint 1) by @orgaralf in #1598
- feat(deploy): package architecture-safe AWS Marketplace assets by @cristiangrxs in #1605
- feat: conversation continuity & deep memory — durable summary + message digests (Sprints 2–5) by @orgaralf in #1603
- fix(deploy): make AMIs ingestible by AWS Marketplace by @cristiangrxs in #1606
- feat(chat): keep conversation files in context for follow-up requests by @orgaralf in #1607
- fix(deploy): require an explicit web ingress CIDR by @cristiangrxs in #1609
- feat: first-run setup wizard for installations without an administrator by @cristiangrxs in #1575
- fix(ci): stop taking the release digest from a cached tag lookup by @cristiangrxs in #1613
- fix(models): roll the gpt-oss-120b sorter default out to existing installs by @orgaralf in #1614
- ci: run frontend-build on the free runner by @cristiangrxs in #1616
- ci: build the arm64 release image on a native runner by @cristiangrxs in #1617
- ci: split the chromium E2E suite into three shards by @cristiangrxs in #1618
- feat(chat): speed-config model mixes in the chat window by @orgaralf in #1620
- feat(chat): recording indicator for record-then-transcribe voice input by @orgaralf in #1621
Full Changelog: v4.3.6...v4.4.0