Changes from Hollama

Llooma is a fork of Hollama by fmaclen. This page is the record of what has been added, reworked or removed since, kept as a whole rather than release by release.

For what shipped in a given version, see the changelog. For what is planned, see the roadmap.

Providers & connections

• Per-connection colour, assigned at creation from the unused swatches, shown wherever that connection’s models appear
• Per-connection model display names, edited in a searchable sub-view of the Servers tab
• Connection card redesigned: status and model count at a glance, everything editable behind one disclosure
• Sync replaces the separate verify/refresh actions, and the verified date is persisted
• Stored API keys now show as “Key saved” instead of an empty field
• New Claude (Anthropic) provider: preset OpenAI-compatible endpoint, just paste an API key
• New Infomaniak provider: Product ID field, endpoint URL built automatically (API v2)
• Provider selection reworked into a card grid (Ollama, OpenAI, Claude, Infomaniak, OpenAI-compatible)
• Identified providers hide their Base URL behind an “Advanced” disclosure (still editable)
• Per-provider “how to get an API key” help links
• Provider metadata centralised in a single PROVIDERS registry

Chat & sessions

• New “Chat” settings tab
• Default model setting (moved out of Interface, renamed for clarity)
• AI-generated session titles: auto-named after the first reply, using a dedicated (cheap) model
• Header shows the conversation title, keeping #id as a parenthesised link
• Copy a whole conversation as JSON or Markdown
• Model picker and per-conversation settings joined into one control in the header
• Message layout reworked: the assistant answers as plain prose, the user speaks in a bubble (optionally tinted with the accent colour), actions sit under the message they act on
• Optional timestamp per message
• Reasoning shown behind a left rule rather than a card inside a card
• Thinking indicator with an elapsed counter past three seconds
• “Scroll to the latest message” button, and a fix to the auto-follow that stopped on a single pixel of rounding
• Bare URLs are linkified and shortened to host/…/page

Data

• Full backup & restore: export/import every data source in a single file
• Shared applyToStore helper de-duplicating the import logic

UI / UX

• Welcome / home page redesign (chat input, suggestion chips, recent sessions)
• Redesigned chat input and model picker
• Settings moved from a /settings route into a modal
• Sidebar redesign with avatar + name
• Theme system: System / Light / Dark modes + Classic, Dracula, Catppuccin, Gruvbox, Nord and Solarized styles, each with its own light and dark ramp
• Every settings tab rebuilt on the same shared components (SettingsPanel / SettingsSection / SettingsField)
• Dialogs go full screen on phones, with a blurred, animated backdrop
• All dropdowns unified on two bits-ui primitives; one model picker used everywhere
• New-user onboarding, and an OIDC-provisioned profile is read-only
• Profile settings (name, avatar, color)
• New logo & favicon

Infrastructure & tooling

• Versioned releases: semantic-release reads the commit log, tags, writes the release notes and publishes ghcr.io/cedhuf/llooma:latest plus a version tag
• In-app update check against the GitHub releases, announced once per version by a dismissible notice (admins only, in server mode)
• CI/CD migrated from npm to pnpm, Node 20 → 26
• Docker image uses GITHUB_TOKEN (no PAT/secrets to manage)
• Added docker-compose.yml and .env.example (configurable port & allowed hosts)
• NixOS OCI container example (nixos/services/hollama.nix)

i18n

• Automatic English fallback: a missing key falls back instead of leaving a blank
• Locales reduced to English and French; adding a language no longer means auditing every key
• pnpm run i18n:status reports coverage

Migrations

• Tailwind v3 → v4
• Svelte 4 → 5 (runes)
• bits-ui v0 → v1
• npm → pnpm

Fixes & cleanup

• Fixed an infinite $effect loop in the About tab (effect_update_depth_exceeded)
• Fixed a <button>-in-<button> hydration mismatch in the model picker
• Removed the prefer-writable-derived anti-pattern across the codebase
• Replaced hacky LocalizedString label casts with proper i18n keys
• Server-side proxy for OpenAI-compatible providers (CORS bypass)
• URL param auto-submit (?q=, &model=)
• Removed Electron