Local Engine Ready
Download Get Started
Local-First AI Workspace

One request. A team of specialists. No trade-offs.

Celestos routes what you say to the system built to do it — chat, a multi-agent team, a scheduled watcher, a browser, an app builder, a video pipeline — all running privately on your machine, with cloud power only when you ask for it. For unambiguous requests, it doesn’t start the model at all — the task runs directly.

Get Started Free See All Features
Request Router
Orb
Hive
Sparks
Browser
Studio
Video
Memory
Skills
Models
Image
Chief of Staff
Mobile
12 systems scanned● match found
No subscription
No API key
Runs fully offline
Data never leaves your machine
No account required
Works without internet
Open model support
MIT & Apache licensed
Private by default
No cloud required
Process // From Words To Result

No menus to learn. Just tell it what you're after.

Every request follows the same three-beat path, whether it ends in an answer, a scheduled watch, or a finished app.

01
Describe
Say what you want, plainly
Type or speak an outcome — "make a TikTok from this video," "watch this price," "build me a landing page." No syntax, no commands to memorize.
02
Route
Celestos picks the system
Your request is scored against all twelve specialist systems and routed — sometimes to one, sometimes to a coordinated combination — then run end to end. For obvious requests, it skips the model entirely and starts the task directly.
03
Deliver
You get the finished thing
A file, a video, a running app, a scheduled alert, an answer with sources — not a transcript you still have to act on.
Fast Path — When Celestos skips the model entirely
For unambiguous requests — a YouTube link with a clear instruction, a price alert, a recording that needs minutes — Celestos detects the intent before the model loads. No warm-up. No wait. The model gets involved when judgment is needed. For everything else, it just goes.
Capabilities // 12 Systems, One Interface

Every request gets routed to whichever system does it best.

Describe an outcome in plain language. Celestos decides which of these twelve systems — or which combination — gets it done. Click any card for the full walkthrough.

01
Orb & Chat
Talk naturally
Two interfaces for talking to Celestos. Orb is your home screen — action-first, with status, a result dock, and next actions. Chat is answer-first — text, links, sources, and a screenshot only when it actually helps.
Try: “Summarise this PDF and list action items”
View feature →
02
Hive
Team collaboration
Multi-role AI teamwork. One request → a manager plans → specialists work → a coordinator checks → you get the result with a live trace.
Try: “Design a brand identity for an eco-friendly coffee startup”
View feature →
03
Sparks
Auto-running tasks
Autonomous tasks that run on a schedule. A Spark wakes, fetches fresh web context, an AI evaluates your condition, and pushes a result only when it matters.
Try: “Alert me when the RTX 5090 drops below $1500”
View feature →
04
Browser
Web automation
Reads pages, searches the web, takes screenshots, fills forms, logs into services. Smart UX: answer-first unless you explicitly ask to see the page.
Try: “Find recent Reddit posts about X and summarize complaints”
View feature →
05
Studio
Prompt to app
Prompt to app. Describe a website or app, pick a template and model, hit generate. Real files stream into an editor with a live preview rendering side by side.
Try: “Design a tea brand landing page with earthy luxury and soft motion”
View feature →
06
Video Toolkit
Create content
Paste a link. Get back a scene-cut, AI-voiced, captioned clip — with title, hooks, and hashtags — formatted for TikTok, Reels, or YouTube. Or point it at any page and it records a scrolling video with voiceover.
Try: “Make a 60-second vertical TikTok from this YouTube link”
View feature →
07
Memory & Projects
Context that sticks
Group related work into Projects so context carries across sessions. Celestos builds a memory graph (Mempalace) from what you feed it — the more you use it, the smarter it gets.
Try: “Start a project for my book writing and dump my notes here”
View feature →
08
Skills & Skill Vault
Auto-activating
Reusable expertise bundles. Skills auto-inject themselves when your message matches them — turning Celestos into a specialist, automatically. Celestos can write its own.
Try: “Create a skill for my specific research methodology”
View feature →
09
Models
Intelligence routing
Pick the brain. You control which model runs which task — local & private or cloud & powerful. Intelligence tiers route heavy reasoning to frontier models and quick steps to fast ones.
Try: “Use local model for privacy-sensitive work, cloud for hard reasoning”
View feature →
10
Image Generation
On-device, in chat
Type /image and a prompt. Celestos generates the image on your machine — no cloud, no subscription, no separate app. The result appears inline in the chat.
Try: “/image a neon cityscape at dusk, cinematic lighting”
View feature →
11
Time-Series & Analytics
Forecasting & SQL
Native time-series forecasting built-in. Hook into live market data or drop a massive CSV, and Celestos natively charts, predicts, and queries it with an embedded analytical database.
Try: “Forecast the S&P 500 for the next 30 days”
View feature →
12
Code & Sandbox
GitHub Auto-Installer
Paste a GitHub link. Celestos automatically detects the language, builds an isolated environment, resolves dependencies, and runs the app on your machine — streaming the live logs back to your chat.
Try: “Clone and run this github repo locally”
View feature →
Also in Celestos // Seven capabilities worth exploring
Multi-panel Workspace — compare builds, models, and tools side by side Cloud Hub — every popular provider in one catalog Benchmarks — know what your machine can run Chief of Staff — proactive briefing engine Mobile Tunnel — Celestos on your phone Mood Engine — AI that reads and predicts your state Dream — self-improving code & skill forge Memos — instant voice, text & file capture Memory Podcasts — your sessions as audio Drop Anything — books, audio, video, PDFs, folders
Routing // Intelligence Tiers

Heavy reasoning goes to frontier models. Quick steps go to fast ones.

Inside Hive, every role is assigned a tier. You decide how much intelligence a task deserves — and what it costs.

Elite
≥ 80 IQ
Frontier reasoning for the hardest thinking — brand strategy, complex synthesis, architecture decisions.
Expert
≥ 65 IQ
Capable and strong — the default for most specialist roles inside a Hive run.
Standard
≥ 50 IQ
Mid-tier and fast — handles most steps in a workflow without slowing the run down.
Basic
< 50 IQ
Smallest and fastest — ideal for edge devices and high-volume, low-stakes steps.
Engine — Live Parameters
Runtime Local — On-Device
Data Residency On-Device
Cloud Providers 18 Supported
API REST
The model runs locally by default, so your data never leaves your machine. Switch any role to a cloud provider from the model picker whenever a task calls for more horsepower — you stay in control of the trade-off.
Workspace // Multi-panel mode

Build three directions. Compare every answer. Keep the whole run visible.

Celestos can run as a single view, split screen, vertical split, three-panel desk, or quad workspace. Put Studio beside Chat, Hive beside Browser, Tracker beside Memory, or compare different LLM outputs while building multiple website versions from one request.

01
Side-by-side builds
Studio + preview
Generate multiple websites or app screens from the same brief, keep their previews open, and choose the strongest direction without losing the prompt trail.
02
LLM comparison
Same task, different brains
Run one request through local and cloud models, compare the outputs, then route future work to the model that actually performs best.
03
Operator desk
Chat + Hive + Browser + Memory
Keep the live conversation, agent trace, browser automation, and memory graph open at once while Celestos works.
LOCAL FIRST
Runs on your machine. No data sent anywhere, ever.
ZERO LOCK-IN
Open models, open license. Yours to keep.
WORKS OFFLINE
No internet. No API key. No account. Just run it.
BUILT FOR POWER
Nine specialist systems. One interface. Your workflow.
Setup // Five Minutes, Start To Finish

Download, start the server, describe what you want.

No account walls, no config wizard. Three steps and you're issuing your first request.

1
Download & install
Grab the installer for Mac, Linux, or Windows and run it like any other app.
2
Start the server
# Mac open -a Celestos # Linux ./celestos # Windows celestos.exe
3
Open it and ask
Visit localhost:3000, drop a file into Orb, and type "summarize this and list action items." That's the whole learning curve.
Full Getting Started Guide Download Page
FAQ // Common Questions

What people ask before they get started.

Yes. The language model runs locally on your machine. Your data stays on your machine unless you explicitly choose a cloud provider.
No. Describe what you want in plain language — "make a TikTok from this video" or "total these invoices" — and Celestos figures out the rest.
A modern CPU with 8GB or more of RAM is recommended (a base model Mac Mini runs it easily). An NVIDIA card with CUDA support speeds things up significantly.
Yes. Switch between local and cloud models from the model picker at any time — you control which one runs which task.
Yes. Celestos has a built-in secure tunnel. Launch it once on your Mac or PC and a mobile-optimised interface becomes accessible from any device — your computer stays the private backend, nothing goes to a third-party server.
Yes. Type /image followed by a prompt in Chat or Orb. Celestos generates the image on your machine — no cloud service, no subscription. The result appears inline in the conversation.
Workflows are end-to-end pipelines triggered by plain language. Say “make a TikTok from this YouTube link” and Celestos handles every step — download, scene detection, transcription, AI voiceover, captions, formatting — and delivers the finished file. There are 70+ workflows covering video, meetings, research, content, and more.
The glowing orb in the bottom-right corner of every page is a separate assistant mode — a preview of the real Orb/Chat interaction model, available wherever you are on the site.
Ready When You Are

Describe the outcome. Celestos does the rest.

Get Started Free View Docs