Aiden

Positioning#

Aiden is a local-first autonomous AI operating system designed to handle desktop GUI operations, browser automation, terminal execution, and unified multi-channel dispatch within a single runtime, addressing the limitation of existing cloud-dependent AI Agents lacking direct local system control.

System-Level Control#

• Desktop Control: nut-js based computer use module with screenshot + OCR, mouse/keyboard automation, and window management.
• Browser Automation: Playwright integration supporting open_browser, browser_click, browser_type, browser_extract, browser_get_url, etc.
• Terminal Control: Shell command execution and code sandbox runtime.

Intelligent Routing & Model Management#

• Multi-Provider Routing: Self-healing provider chain supporting Ollama (local-first), OpenAI, Anthropic, Groq, Cerebras, NVIDIA NIM, OpenRouter, Mistral, and 15+ providers.
• High Availability: keepalive fast cache (<30s responses), exponential backoff retry, automatic failover.
• Daily Budget Control: DAILY_BUDGET_USD env var limits cloud API spend (default $5.00/day).

Tools, Skills & Extensions#

• Built-in Tools: 80+ tools covering web search, file I/O, clipboard, LocalSend LAN transfer, system info, calendar, email (IMAP/SMTP), etc.
• Skills System: 1,400+ composable skills (self-reported count), each as an independent directory (SKILL.md, tools.ts, sandbox.ts); Planner auto-activates by user intent; installable from Registry.
• Plugin Extensions: Drop .js files into workspace/plugins/ to register custom slash commands.

Memory & Reasoning Architecture#

• 6-Layer Memory: Episodic (context window), BM25 keyword (SQLite FTS5), Vector semantic (SQLite + embeddings), Procedural (Skill prompts), Goal (JSON persistence), LESSONS.md (append-only permanent failure log).
• Subagent Swarm: Spawn N parallel sub-agents with voting, merging, and selection mechanisms.
• Identity System: SOUL/USER/GOALS framework, identity refreshed every turn, real-time USER.md editing during conversation.

Interaction & Channel Integration#

• Multi-Channel: Unified Agent Loop for Discord, Slack, Telegram, WhatsApp, Email, Webhook, Twilio, CLI, and Desktop GUI.
• Voice Pipeline: STT (Groq → OpenAI → local Whisper.cpp) + TTS (Edge TTS → ElevenLabs → Windows SAPI), supporting fully offline voice loop.
• MCP Server: Connectable as MCP Server to Claude Desktop and other external clients.
• OpenAI-Compatible API: Base URL http://localhost:4200, Model aiden-3.13, no API key required; compatible with Open WebUI / Chatbox / Cursor.
• Honest Failure: Tool failures report tool name, provider, retry count, fallback chain, and error details.

Architecture Overview#

Channel Input → ChannelAdapter → Planner → AgentLoop
    ├── ToolDispatcher (80+ tools)
    ├── SubagentManager (parallel agents)
    └── MemoryManager (6-layer memory)
        → ProviderRouter (Ollama priority 0, 15+ providers)
        → Responder → stream back to channel

Core at core/agentLoop.ts, event-driven; Desktop GUI on Electron (auto-update); Dashboard on Next.js; Landing Page on Cloudflare Workers.

Installation#

Prerequisite: Node.js 18+

• Zero-install: npx aiden-os
• Global: npm install -g aiden-os && aiden
• Windows installer: irm aiden.taracod.com/install.ps1 | iex
• Linux/macOS: curl -fsSL aiden.taracod.com/install.sh | bash
• Source: Clone, npm install && npm run build && npm start

First run auto-guides Provider selection, API key verification, and config save. Groq offers free quota; Ollama supports fully offline operation.

Use Cases#

• Personal all-purpose local assistant (file management, document summarization, daily briefings)
• Browser automation RPA
• Unified multi-platform bot backend (Discord/Slack/Telegram, etc.)
• Privacy-critical fully offline AI environment
• Local LLM API proxy gateway

Key Constraints#

• Core code under AGPL-3.0; Skills independently under Apache-2.0
• Primary language: TypeScript/JavaScript (Node.js)
• Current version: v3.19.x "ALIVE"
• Windows 10/11: GUI + CLI; Linux/WSL 2/macOS: Headless API + CLI
• Skill count (1,400+) is self-reported, not independently verified
• Project may have been renamed from devos-ai; exact migration history unconfirmed
• Security audit status unknown