Aetheryte

Shared Crystal for AI Agents. Skills, memories, knowledge graph.

23 autonomous agents built this TUI in 90 minutes. Read the metrics.

60-Second Demo

Demo GIF coming soon — clone and run aether --demo to see it live.

npx @agenticide/install
aether --demo

No API keys. No Docker. Just run it.

Install

# One command install
npx @agenticide/install

# Then run
aether

# From source
git clone https://github.com/OWNER/aether.git && cd aether && ./scripts/install.sh

What It Does

• Agent code sessions -- Type prompts, get streaming AI responses, approve tool calls inline
• Fleet orchestration -- /fleet "<prompt>" generates a plan, spawns autonomous agents in parallel, and opens a live per-agent monitoring panel
• Plan mode -- Numbered tasks with category colors and checkboxes; plans persist to PostgreSQL across sessions
• Skills & jobs -- Curated skill sets that make your AI smarter per task; refresh automatically on /switch
• Knowledge graph -- Skills, memories, and relationships, searchable by vector or graph
• Hybrid search -- SearchPipeline detects aide-cloud and auto-captures memories from results
• Overnight monitoring -- /overnight command + status bar shows fleet health across long runs
• Session persistence -- Transcript auto-saved and restored on next launch; no lost context
• Config command -- /config shows and edits model, modeCycle, overnightEnabled, api_url inline
• Toast notifications -- Inline transcript entries with auto-dismiss for non-blocking feedback
• Multi-model -- Ctrl+N cycles models mid-session; locked during active agent execution
• 100% local -- All data stays on your machine. No telemetry. Ever.

Architecture

Three codebases form the crystal:

• aide-cloud -- Hono API + PostgreSQL/pgvector + Ollama embeddings
• Aether (this repo) -- Orchestration, scripts, TUI, skills
• Agenticide -- MCP-native skill jobs (npx @agenticide/install)

Dual-cognition retrieval: Type 1 (fast vector similarity) for instant recall, Type 2 (graph traversal) for complex reasoning chains. See docs/architecture.md for details.

Full Setup

For the full crystal with skills, memories, and fleet:

aether setup

This starts Docker + PostgreSQL + Ollama locally.

| Component | Minimum | GPU Mode | |-----------|---------|----------| | RAM | 8 GB | 16+ GB | | Disk | 5 GB | 20 GB | | GPU | None (CPU) | NVIDIA RTX (10x faster) | | Runtime | Docker, Node 18+, Bun | + Python 3.11 |

Commands

| Command | Description | |---------|-------------| | /help | Full command reference | | /status | System health dashboard | | /profile | Manage skill jobs | | /skills | Browse active skills | | /fleet | Fleet orchestration | | /spawn | Launch an autonomous agent | | /resume | Resume a previous session | | /config | View/change settings | | /dashboard | Open browser dashboard |

Keybindings

| Key | Action | |-----|--------| | Shift+Tab | Cycle modes: Normal → Plan → Haste | | Ctrl+N | Cycle through configured models | | Ctrl+P | Toggle PLAN mode | | Ctrl+F | Fleet widget | | Ctrl+\ | Skills widget | | Ctrl+K | Logs widget | | Ctrl+C | Abort agent / quit (double-tap when idle) | | ? | Keybinding help overlay |

Dashboards

• :5177 -- Aetheryte Command Center (agents, cycles, search, tech debt)
• :5178 -- Overnight Monitor (live event feed, process status)

Contributing

See CONTRIBUTING.md.

License

MIT