Aether

Autonomous Engine for Theoretical & Hands-on Exploration in Research

An AI research assistant for researchers, built on OpenCode.

🌐 aether.aiphys.cn

English | 简体中文 | 繁體中文

---

Overview

Aether is a feature-rich AI research assistant platform. It runs with a client/server architecture — the CLI starts a local HTTP server, providing a full interactive interface through the browser. Mobile access (WeChat / Feishu / QQ) is also supported. Desktop client (Electron) support is planned for the future.

Key Features

• Ready to use: Download and double-click to launch; built-in default model settings
• Full coding capabilities: Code execution, LSP code intelligence, file and terminal operations — covering the entire daily development workflow
• 25+ AI providers: Supports Anthropic, OpenAI, Google Gemini, AIhubmix, DeepSeek, Z.AI, Kimi, Qwen, and other mainstream platforms, plus any OpenAI-compatible API
• 10+ built-in research Skills: Literature review, paper writing, deep research, arXiv search, peer review, research grant writing, and more — ready to use out of the box
• Skill self-evolution: After completing a task, the Agent automatically reviews the conversation history and solidifies successful experiences into reusable Skills (copy-on-write writes, security scanning, version snapshots), forming a continuous learning loop
• MCP protocol support: Integrates Model Context Protocol — connect local or remote MCP servers to extend the toolset
• Knowledge base (RAG): Vector-index PDF and text documents, with semantic search that injects only relevant context, significantly reducing token usage
• PDF reading mode: Read PDFs directly in the Web UI with highlights, bookmarks, notes, and AI-assisted translation and Q&A
• PDF to Markdown: AI-driven multi-stage pipeline that converts PDF papers into high-quality editable Markdown (with formula and image extraction)
• Markdown translation: AI-powered Markdown document translation that automatically preserves LaTeX formulas
• Voice input: Speech-to-text powered by multimodal models, automatically removing filler words and correcting technical terms
• Git integration: View branches, commit history, diffs, and file changes in the UI
• Scheduled tasks: Supports cron expressions, fixed intervals, and one-time scheduled tasks for automating research workflows
• Memory mechanism: AI automatically persists user preferences and interaction experiences to memory.instruction.md, reused across sessions without repetition
• Session sharing: Generate shareable links with real-time conversation sync

---

Installation & Launch

Installer Script (Recommended)

Download the installer script for your platform from the official website for one-click installation with automatic updates. After installation, simply launch from your system applications.

Manual Installation

Download the archive for your platform from the Releases page.

Web Browser Version

Extracted directory layout:

aether          ← CLI binary (aether.exe on Windows)
web/            ← Frontend static assets (must stay alongside the binary)
Aether.vbs      ← Windows launcher
Aether.command  ← macOS launcher

Windows: Double-click Aether.vbs (if it errors, run Aether.exe first) — the browser opens automatically.

macOS: Grant execute permission on first use, then double-click Aether.command to launch:

chmod +x Aether.command aether   # first time only

Linux:

chmod +x aether   # first time only
./aether web

After aether web starts, it displays local and network access URLs, and the browser opens automatically.

Electron Desktop Version

Coming soon

Running from Source

For development and debugging. Requires: Bun 1.3+

bun install

Web UI Development Mode

Requires two terminals to start the backend and frontend separately:

# Terminal 1: start the API server
bun dev serve

# Terminal 2: start the frontend (then open the displayed http://localhost:xxxx)
bun run --cwd packages/app dev

If resources fail to load after the page opens, manually add the http://localhost:xxxx URL shown in Terminal 1 to the server list.

---

Configuring AI Providers

You can use the built-in free providers, or add AI providers through the settings UI (recommended).

Feature Details

Built-in Skills

Aether includes 16 research-oriented Skills that auto-trigger based on their descriptions — no manual invocation needed:

Category	Skill	Function
Dev Tools	skill-creator	Create and optimize Agent Skills
	skill-manager	Scan, classify, and manage Skills collections
	code-reviewer	Code review (security, performance, best practices)
	project-signpost	Generate project directory navigation files
Academic Research	academic-researcher	Literature review, paper analysis, equation derivation
	literature-review	Multi-database systematic literature review
	peer-review	Peer review toolkit
	write-paper	Generate LaTeX papers from research project files
	deep-research	Multi-source synthesis and deep research
	arxiv-search	Search arXiv for preprints
	read-arxiv-paper	Read and analyze arXiv papers
	research-grants	Write research grant proposals
	scientific-critical-thinking	Scientific evidence quality assessment
	scientific-brainstorming	Research hypothesis generation and interdisciplinary exploration
	response-to-referee	Point-by-point responses to reviewer comments
General	brainstorming	Turn ideas into designs before implementation

Academic Skills can be chained: arxiv-search → read-arxiv-paper → literature-review → write-paper

Knowledge Base

Vector-index PDF papers or text documents (.md, .txt, .json, .yaml, .csv, .tex) in bulk:

• Three embedding options: OpenAI API, local model all-MiniLM-L6-v2 (no API key needed), or custom endpoints
• Semantic search: Cosine similarity-based retrieval of the most relevant passages, injecting only necessary context rather than entire documents
• Auto-sync: Detects file additions, modifications, and deletions, automatically updating the index

PDF Reading Mode

Read PDFs directly in the Web UI:

• Annotation system: Highlights (four colors), bookmarks, notes
• AI-assisted translation: English-to-Chinese translation that preserves technical terms
• AI Q&A: Answer questions based on the current page and surrounding context pages
• First-read summary: Automatically generates a full-document summary when opening a document

PDF to Markdown

AI-driven multi-stage pipeline:

1. Page rendering → text extraction → figure localization → formula recognition
2. LaTeX syntax validation and content integrity checks
3. Automated fixes and cross-page quality verification

MCP Tool Integration

Connect external MCP (Model Context Protocol) servers to extend AI capabilities:

aether mcp add <name> <command>    # Add a local MCP server
aether mcp list                     # List configured servers
aether mcp auth <name>              # OAuth authentication
aether mcp debug                    # Debug MCP connections

Mobile Access

Chat with Aether from your phone via WeChat, Feishu, or QQ bots. Configure credentials for the respective platforms in the Web UI settings.

---

Other Commands

aether web            # Start web server and open browser
aether serve          # Start headless API server
aether run <message>  # Non-interactive command execution
aether models         # List available models
aether providers      # View providers
aether mcp            # Manage MCP servers
aether session        # Session management
aether stats          # View usage statistics
aether upgrade        # Self-update
aether debug config   # View current configuration

---

Troubleshooting

API returns 404: Check that baseURL ends with /v1.

API returns 429: Wrong key, or overridden by a stale auth.json — delete ~/.local/share/opencode/auth.json and retry.

Browser doesn't open automatically: Visit the URL shown in the terminal manually (requires xdg-open).

Frontend assets not found: Ensure the aether binary and web/ directory are in the same folder.

Model list is empty: Check that the OPENCODE_CONFIG environment variable path is correct.