v0.5.0
✦ Your stories, written among the stars ✦
The biggest update yet: a lorebook system that reads with you, phrase bans that
scrub the AI-isms out of your prose, variant takes, image input, bookmarks,
folders — and a night sky that reacts to the colors in your writing.
🏰 Lorebooks — worlds that stay consistent
- Build a collection of titled lorebooks and enable them per chat. New chats start
with lore off — nothing gets injected unless you ask. - Paste a whole world bible into one entry: smart retrieval finds and sends
only the passages your current scene needs — a 300 KB encyclopedia sends a
few paragraphs, not the whole thing. A 🌍 note on each reply shows exactly
what was pulled. - Trigger-word entries for facts you always want whole when a word appears,
and always-include entries for core canon. - Optional semantic matching (off by default): flip it on and retrieval
matches by meaning — "the scaled beast" finds your dragon lore — using a
small model that runs entirely on your machine. - Attach
.md/.txtfiles to entries or to your system/project instructions
as living context.
✍️ Writing tools
- Phrase bans — list the words models overuse ("delve into", "a tapestry
of") and Constellation quietly swaps or strips them after generation. The
model never sees the list, so it never leans into it. - Regenerate variants — ↻ now keeps every take. Flip between them with
‹ 1/3 › and continue from the one you like. - Bookmarks (★) — star any passage in any chat and jump back to it later.
- Craft coach improvements and a growing craft journal.
🌌 A sky that listens
- Color words in a reply now light up the background as you scroll past them —
and optional cosmic events (comets, planets, supernovas, flaring stars)
bloom in that color at the edges of your screen. - Fully controllable: Settings → Color atmosphere (intensity, reach,
spread, blend, events on/off, event size).
🧰 Everything else
- Image input for vision models (attach or drag-drop; switch to a vision
model likeglm-4.5vand the general endpoint) - Folders to group chats, pinned chats, full-text search with jump-to-match
- Per-chat draft autosave, usage/token tracking, export to Markdown
- Preserved Thinking carried between turns · thinking indicator while replies
warm up · scroll that stays put while you read - One-file backup & restore now covers lorebooks, bookmarks, phrase bans
- glm-5.3 support
- New README with a proper tour — and yes, it's all AI-generated code.
⚠️ Upgrade notes
- Your data is safe — updating never touches your chats or settings in
%AppData%\constellation. A backup before updating never hurts, though. - The installer is bigger now (~170 MB) because the optional on-device
semantic-matching engine is bundled. If you never enable semantic matching,
it never runs and never downloads anything. - If you had a lorebook before: it migrates to a titled lorebook called
"Main" on first launch — entries intact. Existing chats now have lore
off by default; open ❖ and tick "Main" to re-enable per chat. - Vision needs the general endpoint (not the Coding Plan endpoint).
Made with GLM. The sky is yours. ✦