A private, local-first writing companion for roleplay, fiction, and craft — set against a night sky that remembers your story.

Yes this is all AI generated code, I don't know how to code. What do you want from me?

---

Constellation is a desktop home for long-form AI-assisted writing. Bring your own key — GLM (Zhipu AI / Z.ai) out of the box, or any OpenAI-compatible provider (OpenRouter, Ollama, LM Studio…) — and get a calm, focused space built for stories, not chat transcripts: a lorebook that only sends the passages your scene needs, a Chronicle that keeps you from misremembering your own plot, variant takes you can flip between, and phrase bans that quietly scrub the AI-isms out of the prose. Everything — chats, lore, settings — lives in plain files on your machine.

It's a client: Constellation has no AI of its own and no accounts. Your key talks straight to the endpoint you configure; your writing never touches anyone else's server.

---

✦ Why you might want it

📜 A Chronicle for the reader, not the model. Fifty messages deep, you're the one who misremembers — who knew what, who's injured, what was promised. Open the Chronicle panel and distill your story into durable facts with one click. It reads in quiet chunks with a live progress bar, only processes what's new since the last pass, and every fact is editable and deletable. It's your reference; nothing from it is ever sent to the model.

🌌 Your stories, written in the sky. Every chat owns a constellation: one star for the story's first light, one more for every moment you bookmark. The current story's pattern hangs beside your prose — hover a star to glimpse its moment, click to travel back to it. Open the Sky map and all your stories hang in one sky together. And the sky weathers your story: tension red-shifts the stars and makes them flicker, sorrow dims and slows them, joy warms them gold.

🏰 Worlds that stay consistent. A lorebook holds your world bible — a whole 300 KB encyclopedia if you like — and only the passages relevant to the current scene are sent to the model. Entries fire on trigger words, or let smart retrieval find the right paragraph on its own. Enable a lorebook per chat, so your Warhammer campaign and your romance novella never contaminate each other. When lore feeds a reply, its star pattern lights up in the margin (Living Constellations) — click it to see exactly what the model was given.

✍️ Prose without the tells. A phrase-ban list quietly swaps or strips the words models overuse ("delve into", "a tapestry of") after generation — the model never sees the list, so it never leans into it. And when the story mentions colors, the words themselves wear them — 223 named colors, readable even for obsidian and onyx.

🎚️ Take the best take. Regenerate variants keeps every version of a reply — flip between them with ‹ 1/3 › and continue from the one you like. Fork any message into a branching chat. Bookmark lines worth coming back to.

🔌 Any provider, your key. GLM's endpoints are built in (including Coding Plan keys), or point Constellation at OpenRouter, Ollama, LM Studio, or any OpenAI-compatible base URL with a custom model ID. Thinking effort carries over where the provider supports it.

🔒 Private and portable. No cloud, no accounts, no telemetry. Chats, lorebooks, chronicles, presets and journals are plain JSON/text in %AppData%\constellation — back up and restore the whole app to one file. The renderer is sandboxed and every IPC surface is validated.

🖼️ And the practical bits. Image input (vision models), collapsible preserved thinking, folders, pinned and hidden chats, full-text search, drafts, export to Markdown, a writing coach with a craft journal, zen mode, and a token meter.

---

🚀 Quick start

You need two things: the installer and an API key.

1. Download Constellation Setup 0.6.0.exe from Releases and run it. No Node.js, no command line — the runtime is bundled.
2. Launch Constellation, click the ✦ star (top-left) → Connection.
3. Paste your API key and pick the matching endpoint:
   • GLM — Coding plan (api.z.ai) — for GLM Coding Plan keys (most common)
   • GLM — General (api.z.ai / bigmodel.cn) — standard GLM keys
   • OpenRouter — or Custom for any OpenAI-compatible provider (Ollama, LM Studio…)
4. Save — the status should read connected. Type something and press Enter.

💡 Wrong endpoint is the #1 cause of 429 errors. If you're on a Coding Plan key, use the Coding Plan endpoint.

First things to try

• Ask for a vivid description of a sunset, then scroll through it slowly — watch the words color and the margins glow. 🌌
• Bookmark a line you love (★) — then look right: a star was just born in your story's constellation. Click the pattern for the Sky map.
• Open ❖ Lorebook → create one called My World → paste your setting notes → tick it on for your chat.
• A few exchanges in, pull the CHRONICLE tab on the right edge → ✦ Capture new.
• Settings → Phrase bans → add delve into = explore and watch it vanish from replies.

---

✨ Feature tour

Chat	Streamed markdown replies · expandable code blocks · edit & resend · stop mid-stream · continue after length-cuts · copy anything
Thinking	Collapsible reasoning blocks, saved with the chat; models carry reasoning between turns (Preserved Thinking)
Chronicle	Per-chat fact panel for the reader · chunked capture with live progress · incremental (reads only what's new) · Rebuild re-reads · editable facts · never injected into prompts
Story Constellations	Every chat is a constellation · stars = bookmarks · margin pattern with hover previews + jump · Sky map of all stories
Lorebooks	Titled collection · per-chat enable/disable · trigger-word entries · smart passage retrieval (BM25 + optional on-device semantic matching via nomic embeddings) · attach files to entries · 🌍 log + Living Constellations show exactly what was pulled
Instructions	Per-chat system + project prompts · presets · file attachments as context · first-run template files you can rewrite freely
Writing tools	Phrase bans & substitutions · variant takes ‹ n/m › · forks with lineage links · bookmarks (★) · craft coach + journal
Media	Image input for vision models (incl. glm-5.3-flash) · thumbnails in chat · export to Markdown
Connection	GLM (Coding plan / General) · OpenRouter · any OpenAI-compatible base URL + custom model ID · thinking effort mapped for non-GLM providers
Organize	Folders · pinned chats · hidden chats · full-text search with jump-to-match · drafts per chat · usage/token tracking per chat
The sky	Color-word tints in prose (223 names) + margin glows + cosmic events (toggleable, size/blend/reach controls) · mood-weather starfield · parallax · twinkle & density controls · zen mode
Data	One-file backup & restore · everything in plain files under %AppData%\constellation
For tinkerers	Opt-in localhost test server + read-only cli.js for headless poking · tweak theme/star colors in plain CSS

Keyboard: Enter send · Shift+Enter newline · Esc close panel / exit Zen · ? cheat sheet.

---

🔒 Privacy & where your stuff lives

Constellation is a client around your key. Nothing is sent anywhere except your prompts to the endpoint you configure. The optional Chronicle capture sends your story text to your configured model for fact extraction — the results stay local, editable, and are never fed back into chats.

%AppData%\constellation\
├── config\    settings.txt, phrase_bans.txt, project.txt, modes\ (your prompts)
└── data\      sessions\ (chats), lorebooks.json, chronicle.json, presets\,
               bookmarks.json, craft_journal.txt, drafts.json, folders.json

Copy that folder to back up; delete it to fully reset. The in-app Backup button bundles everything into one JSON file. Semantic matching (optional, off by default) runs a local embedding model — your lore never leaves the machine for it.

---

🛠️ Developers — run from source

git clone https://github.com/Notsurejr/Constellation.git
cd Constellation
npm install
npm start        # run from source
npm run dist     # build the Windows installer into dist\

Stack: Electron + vanilla JS (no build step), marked for markdown, @huggingface/transformers for optional on-device embeddings, Google Literata (SIL OFL 1.1) bundled for type. See BUILD_SPEC.md for the full architecture reference and CHANGELOG.md for release history.

Tweaking without code

File	Controls
config\modes\*.txt	The system prompt templates — rewrite freely
src\styles\theme.css	Colors, fonts, star color — the whole look (CSS variables)
build\icon-render.js	Params for regenerating the app icon

---

🆘 Troubleshooting

• 429 / out of credits: almost always the wrong endpoint — Coding Plan keys need the Coding Plan endpoint.
• 401 / invalid key: re-paste the key in Settings → Connection.
• Reply not appearing: check the status light (top-right) and the model/endpoint in Settings.
• Custom provider not working: double-check the base URL includes the version path (e.g. /v1) and that the model ID matches what the provider expects.
• Old icon after rebuild: Windows caches icons — reinstall or refresh the folder.

---

📄 Credits & licenses

Constellation is MIT-licensed. It bundles MIT/Apache-2.0/BSD-licensed libraries and the Google Literata typeface (SIL OFL 1.1) — full inventory in THIRD_PARTY_NOTICES.md. The optional semantic-matching feature downloads nomic-embed-text-v1.5 (Apache-2.0, © Nomic AI) at runtime; the model itself is not bundled.

---

Made with GLM. The sky is yours. ✦