Documentation — Celestra AI

01Getting started

Open Celestra and sign in with Google or email & password.
Pick a model from the model picker. The default is Qwen 3.7 Flash (fast, free).
Chat. Everything else — voice, agents, plugins — works from the same screen.

Chats are stored on your device first and synced to your account. Nothing is shared with third parties beyond the AI providers needed to answer you — see the Privacy Policy.

02Chat commands

Type these in the message box instead of a normal message:

Command	What it does
`/agent <goal>`	Starts Agent Mode for the goal — it plans and calls tools across up to 12 rounds until done. Keep chatting normally while it works.
`/end`	Stops the running agent immediately.

Stop button: the stop icon during streaming cancels the run cleanly — including a 3D generation already in progress.

03Models & providers

Celestra routes to multiple providers with automatic fallback — if one backend fails, the next model takes over and tells you.

CodeCraft (33 models): GPT-5.6 Sol / Terra / Luna, GPT-5.5 & GPT-5.5 Pro, Claude Opus 4.6 / 4.7 / 4.8 / 5, Claude Sonnet 5, DeepSeek V4 Pro / Flash, Qwen 3.7 Max / Flash / 3.8, Grok 4.5, MiniMax, Kimi, GLM and more — many with 1M context.
Free tier (xkiro): Qwen, Mistral, MiniMax, GLM, and other free models — no payment needed.
Switching: open the model picker any time; the choice applies to your next message, mid-chat included.
Vision models get your image attachments; non-vision models get the images stripped automatically instead of erroring.

04Plugins

Plugin	Default	What it gives the model
Web Search	Connected	Live web results for current events and facts
Weather	Connected	Forecasts by location or city
Compute	Connected	Safe math / code evaluation
Image & 3D	On demand	Generates images and 3D models in chat

Each plugin action asks for permission before first execution: Allow once, Always, or Deny. “Always” is remembered per plugin + action. If you decline, the model is told the action was denied — it never pretends it ran.

05Celestra Plus

Plus is a membership that unlocks the premium tier. The app checks your account live, so a grant applies instantly — no update, no restart.

What Plus unlocks

Premium model tier: Claude Opus class models, GPT-5.5 Pro reasoning, and other top-tier CodeCraft routes.
Premium 3D generation lane (three.ws premium) instead of the free lane.
A Plus welcome note in chat the first time it activates.

How do I get it?

Plus is granted manually by the Celestra team against your account email — if you were promised Plus, contact avirallal11@gmail.com with your sign-in email. It appears in the app within seconds of being granted.

Plus status lives on your account, not your device — reinstalling the app keeps it.

06Your account & data

Sign-in: Google or email/password (Firebase Auth).
What's stored: your profile, chats, and Plus status (see the Privacy Policy for the full list).
Delete your account: the Delete Account page permanently removes your profile, chats, and sign-in — immediately, no email required.
Memory-only mode: memories and preferences are kept on-device and are never pushed to the cloud on their own.

07FAQ

Do I need to pay for anything?

No. The free model tier works without payment. Plus is optional and granted manually.

Why did my message fail and switch models?

A provider was down or rate-limited. Celestra automatically retries on a fallback model and continues the conversation.

Why does an action ask permission even though I'm the one who asked?

First execution of each plugin action requires consent. Tap Always and it won't ask again.

Can I stop a long run?

Yes — the stop button in chat, /end, or the Stop action in the notification all cancel cleanly, including an in-flight 3D generation.

Which Android versions are supported?

Android 7.0 and newer.

How to use Celestra AI