OBJECTIVE-FIRST AI OPERATING SYSTEM
Restoring workspace
I°
Loading...
Your complete guide to the AI operating system built around memory, skills, and MCP. 50+ apps, persistent memory, multi-agent orchestration, and an AI that evolves with you.
Open Aya OS works in any modern browser — Chrome, Firefox, Safari, Edge. No installation needed. Visit the URL and you are ready to work.
Aya appears in the bottom right corner. Click the Talk button or say a command to start. She responds with voice and text simultaneously — no lag between what you hear and what you read.
Click the central Apps button or say "Show apps" to browse all tools: Talk, Code Lab, Study Helper, Image Studio, and more — all voice-accessible.
Aya is an MCP server and host. Open Developer Hub and go to the MCP tab to see her exposed tools (memory search, graph query, context get) and connect external MCP servers via AYA_MCP_SERVERS.
Aya responds to natural language. Short commands route to the speed tier (DiffusionGemma, 1107 tok/s) automatically — you get near-instant responses for quick turns without any configuration.
Aya listens for commands and executes actions. Short turns automatically route to DiffusionGemma (4x faster) for near-instant responses. Longer reasoning turns route to Claude Sonnet or Opus automatically.
Full conversational AI with sentence-by-sentence streaming TTS — Aya starts speaking after the first sentence arrives, not after the full response is done. Text and audio are always in sync.
Click the microphone icon to switch to dictation. Aya transcribes your speech without executing commands. Perfect for drafting notes, emails, and documents.
Aya proposes single-variable mutations of her own system prompt, A/B tests candidate vs champion on the same eval corpus, and promotes only on a meaningful win. The result compounds over time.
Aya implements the Model Context Protocol (MCP 2025-03-26) as both a server and a host. This positions her as the workspace, memory, permission, and decision layer above MCP-connected models and tools.
Connect Claude Code, Cursor, or any MCP client to /api/mcp. Aya exposes 6 tools and 3 resources.
Set AYA_MCP_SERVERS to a JSON array of server configs. Aya discovers tools from each server and merges them into every reasoning turn.
[
{ "id": "fs",
"name": "Filesystem",
"url": "http://localhost:3001" }
]Quick connect: Add a .mcp.json at your project root and point it to http://localhost:3000/api/mcp. Claude Code and Cursor auto-discover it on startup. See the Developer Hub MCP tab for the copy-paste config.
Aya learns from every interaction and remembers across sessions. Preferences, facts, decisions, and patterns are automatically extracted and recalled when relevant. The discovery gate prevents duplicate memory writes by running a semantic similarity check before storing anything new.
Commands under 10 words automatically route to DiffusionGemma (1107 tok/s vs 303 tok/s for standard models). You get near-instant voice responses for greetings, quick commands, and navigation without any configuration. Longer reasoning turns automatically escalate to Claude Sonnet or Opus.
Aya starts speaking after the first sentence arrives, not after the full response finishes. Text and audio advance in lock-step — what you see in the chat bubble matches exactly what you hear, sentence by sentence.
Visit /research to see Aya's self-improvement loop. She proposes mutations of her own system prompt, A/B tests them on the eval corpus, and promotes only when the candidate wins on mean score with a regression guard. The champion prompt updates automatically.
App open, close, message send and receive, voice start and stop all have distinct synthesised tones (no audio files — generated via Web Audio API). Respects prefers-reduced-motion and can be disabled via the Developer Hub.
Don't memorize commands. Just speak naturally: "I need help with my math homework" or "Can you make me an image of a sunset?" Aya understands intent and routes to the right agent and model automatically.
Open Aya OS includes curated learning paths from top GitHub repositories and supports active recall, spaced repetition, and deep-dive document analysis.
Master data structures and algorithms with real patterns, not random problems.
Complete ML engineering syllabus covering neural networks, computer vision, and NLP.
Learn modern web development with React, Node.js, testing, and DevOps practices.
Just ask Aya. Say "How do I..." followed by what you want to learn. She will guide you through any feature, and if she crystallizes a new skill in the process, it persists for next time.