The AI workspace that runs on your machine.
Code with an AI agent, run local LLMs, research any field and build with friends. One app for Windows.
- Works offline
- Local or cloud models
- Zero telemetry
- You approve every change
- llama.cpp
- OpenRouter
- Hugging Face
- Together AI
- Groq
- Fireworks AI
- DeepInfra
- Cerebras
- Claude Code
- Codex
Five AI tools. One window.
Code with an agent. Keep every decision.
Chat, edit and run in one IDE. Every change waits for your review.
- Ask, Agent and Plan modes
- Built-in terminal
- MCP, skills and plugins
Any model. Your GPU or the cloud.
Run open models on your own GPU, bring your own key, or use the plans you already pay for.
- llama.cpp on your GPU
- 7 open-model hosts
- Claude Code and Codex
Map a whole field. Find the gaps.
Collect a subject’s papers from OpenAlex and arXiv, follow the citations, see what is missing.
- Citation map
- Gap finder
- Daily or weekly watch
Meet Sinai, the assistant with a face.
Talk to Sinai through any model you choose.
- Speaks its answers
- Microphone always shown
- Design its face
Build with your people.
Add friends, see who is online and chat, in the app and on the web.
- Live presence
- One-to-one chat
- Email never shown
Friends: who is online, requests, and one-to-one chat. This panel needs JavaScript.
See it in action.
Powerful agents. You hold the keys.
Review every change
Keep or undo the agent’s edits by part, by file or all at once.
Approve before it acts
Commands, tools, posts and purchases ask you first.
One-key stop
Ctrl+Alt+End stops every running turn, from any program.
Works offline
Local models, your files, the IDE and saved research.
Zero telemetry
No usage analytics, no tracking.
Keys in Windows
API keys live in Windows Credential Manager.
Sealed sessions
“Stay signed in” is sealed to your Windows user.
Nothing types for you
A model never types in your terminal or presses Save.
No install. Code with AI in your browser.
The same chat and editor in a tab, with llama.cpp on your computer or your own API key.
- Open local folders
- Bring your own key
- Review every edit
Opening a folder needs Chrome or Edge. A key you paste stays in the tab and goes only to the provider you chose.
Friends: who is online, requests, and one-to-one chat. This panel needs JavaScript.
Build with your people.
Add friends, see who is online and chat, in the app and right here.
The panel is a demo with sample people, in your browser. Friends needs an Alelyon account; new messages are sealed on your devices before they reach Alelyon; an independent review is pending.
Open source
Open source. Open results.
The verifier and the sign-in and friends client are public, and the lab publishes its misses beside its wins.
A receipt from the lab’s stack can be replayed against your own inputs with a public key obtained out of band. It detects revision of committed inputs, not whether the producer invented them.
Get started
Start free. Stay in control.
No sign-in to use what runs on your computer. An account adds friends and messages.
Without an account, everything in the app runs locally, and the public alelyon-os toolkit needs no account either.
Latest from the lab
All research- Separating model-authored programs from computed resultsThe restricted DSL parses an admitted program and computes its result through an interpreter instead of accepting a recalled figure.
- Evidence required before applying a receipt in another domainA new domain application needs a defined user decision, a covered uncertainty term, and evidence that the receipt helps with that decision.
- The scope of a storage-error execution boundThe execution certificate evaluates supported computations under a declared storage-quantization model and refuses unsupported cases.
FAQ
Questions, answered.
Is Alelyon free?
Nothing on this site asks for payment: the app and an Alelyon account cost nothing today, and no price has been set for later releases. This page will say so before one applies.
Models you bring can cost money through their providers, or through your Claude or ChatGPT plan when you use Claude Code or Codex.
Does it work offline? Do I need an account?
No account is needed. Choose “Use offline” at sign-in and everything on your computer works: your models, files, the IDE and saved research. An account adds friends and messages; sign in with GitHub, Google, Hugging Face or LinkedIn.
Is my code private? What leaves my computer?
Only what you send to a service you chose: a prompt to a hosted model, a search to OpenAlex or arXiv, a message to a friend, a sign-in to Alelyon. On its own, the app only asks Alelyon’s sign-in service whether it is up, with no account details. There is no analytics or telemetry, and with a local model through llama.cpp your prompts never leave your computer.
Can the agent change files or run commands on its own?
Not without you. Its edits wait for your review, commands and tools ask first unless you always allowed that exact one, and Ctrl+Alt+End stops every turn. Auto mode, which only you can switch on, lets the agent use your desktop; purchases, payments, sign-ins and account changes still ask, and it never types a password.
Can I run local LLMs? Which models work?
Yes: GGUF models on your own GPU through llama.cpp. Also open models hosted by OpenRouter, Hugging Face, Together AI, Groq, Fireworks AI, DeepInfra or Cerebras with your key; Claude Code and Codex on your own plans; and any OpenAI-compatible endpoint.
Availability is not correctness. A caught exception, a non-empty dataframe, a bound port, or an HTTP 200 is not evidence that a result is right.