This is an educated lobster. You can also call him AC ( not the AC that rejects your papers )

中文  •   NanoClaw  •  

Features

Compared to NanoClaw, we've primarily modified communication with Feishu etc., memory system, and academic-specific skills:

• Semantic Memory Search - Hybrid SQLite + LanceDB architecture for intelligent context retrieval
• Multi-channel Support - WhatsApp, Telegram, Slack, Feishu/Lark
• Agent Groups - Multi-agent collaboration for complex tasks
• Scheduled Tasks - Automated research summaries and periodic jobs
• Container Isolation - Secure sandboxed execution per group
• Academic Integration - Optimized for research workflows

Academic Skills

AcademiClaw includes 16 built-in academic-specific skills:

📚 Paper Management

Skill	Description	Triggers
zotero-local	Local Zotero database management: CRUD, PDF reading	"zotero", "literature", "paper"
zotero-paper-reader	Read and analyze academic papers from Zotero	"read paper", "analyze paper"
arxiv-fetcher	Fetch papers from arXiv	"arxiv", "fetch papers"
paper-fetcher	arXiv auto-search, translation, Zotero import	"search papers", "检索论文"
paper-summarizer	Generate daily/weekly/monthly reports	"daily summary", "学习总结"
pdf-reader	Deep PDF reading and analysis	"read PDF", "阅读PDF"
papervault-cron	Complete paper automation workflow	"paper workflow"

🧠 Learning & Research

Skill	Description	Triggers
learning-reflector	Deep reflection, blind spot discovery	"reflect learning", "反思学习"
deep-research	Enterprise-grade research: multi-source synthesis	"deep research", "comprehensive analysis"
content-research-writer	Content writing: research-backed, citations	"write", "content"
data-analysis	Data analysis: statistics, visualization	"analyze data", "data analysis"

🛠️ Productivity Tools

Skill	Description	Triggers
token-usage-tracker	Track API token consumption	"token usage", "API消耗"
diagram-generator	Generate flowcharts, sequence diagrams	"draw flowchart", "画流程图"
slidev-generator	Auto-generate Slidev presentations	"create presentation", "创建演示"
github-cli	GitHub automation (repos, issues, PRs)	"create repo", "创建仓库"
agent-browser	Browser automation for research	"browse", "search web"

Usage Examples

@Andy use zotero-local to search papers on machine learning
@Andy search today's papers on transformer    # paper-fetcher
@Andy generate weekly learning summary        # paper-summarizer
@Andy reflect on my learning blind spots      # learning-reflector
@Andy read this PDF paper                     # pdf-reader
@Andy draw a microservice architecture diagram    # diagram-generator
@Andy create a presentation on RAG            # slidev-generator
@Andy create a new GitHub repository          # github-cli
@Andy check my token usage today              # token-usage-tracker

Quick Start

git clone https://github.com//academiclaw.git
cd academiclaw
claude

Initial Setup

Run /setup in Claude Code CLI to configure:

• Dependencies and containers
• Message channels (WhatsApp, Telegram, etc.)
• Memory system (optional)

Memory System

AcademiClaw includes an optional semantic memory system:

Mode	Description	Cost
Off	SQLite only (exact match)	Free
Mock	Test mode, no API required	Free
Jina AI	Quality embeddings	1M/month free
OpenAI	Official embeddings	Pay-per-use

Enable with MEMORY_ENABLED=true in .env.

Usage

Talk to your assistant with the trigger word (default: @Andy):

@Andy summarize the latest papers on semantic search from arXiv
@Andy create a weekly summary of my research notes every Monday 9am
@Andy search my conversation history for "vector database benchmarks"

Architecture

Channels → SQLite + LanceDB → Polling → Container (Claude Agent) → Response

Key components:

• src/index.ts - Main orchestrator
• src/memory/ - Semantic search system
• src/channels/ - Channel implementations
• src/container-runner.ts - Container execution
• src/task-scheduler.ts - Scheduled tasks

Memory System

      ┌────────────┐
      │   Input    │
      └─────┬──────┘
            │
   ┌────────┴─────────┐
   ▼                  ▼
 SQLite           LanceDB
(Source DB)     (Semantic Index)
   │                  │
   └────────┬─────────┘
            ▼
      Hybrid Retriever

Configuration

Key environment variables:

# Memory System
MEMORY_ENABLED=false                    # Enable semantic search
EMBEDDING_API_KEY=                      # Jina AI or OpenAI key
EMBEDDING_BASE_URL=https://api.jina.ai/v1
EMBEDDING_MODEL=jina-embeddings-v3

# Claude API
ANTHROPIC_API_KEY=sk-ant-...
CLAUDE_CODE_OAUTH_TOKEN=                # Alternative

# Assistant Configuration
ASSISTANT_NAME=Andy
ASSISTANT_HAS_OWN_NUMBER=false

See .env.example for all options.

Requirements

• macOS or Linux
• Node.js 20+
• Claude Code
• Docker or Apple Container

Development

npm run dev          # Run with hot reload
npm run build        # Compile TypeScript
npm test             # Run tests

Memory System Tests

# Unit tests
npm test

# Integration tests
npx tsx scripts/test-memory-integration.ts

# E2E tests (with mock embeddings)
MEMORY_ENABLED=true MEMORY_USE_MOCK_EMBEDDINGS=true npx tsx scripts/test-memory-e2e.ts

# Jina AI test (requires API key)
JINA_API_KEY=your-key npx tsx scripts/test-jina-memory.ts

Documentation

• Memory Architecture - Detailed system design
• Test Results - Test coverage
• Requirements - Architecture decisions
• NanoClaw SPEC - Base system specification

Troubleshooting

Memory not working:

• Check MEMORY_ENABLED=true in .env
• Verify API key is set
• Check logs: tail -f logs/academiclaw.log

Container fails to start:

• Ensure Docker is running: docker info
• Check container logs in groups/main/logs/

No response to messages:

• Verify trigger pattern (default: @Andy)
• Check channel credentials in .env

Contributing

AcademiClaw is a minimal fork of NanoClaw. Contributions should follow the same principles:

1. Skills over features - Use .claude/skills/ for add-ons
2. Keep it minimal - Don't add bloat
3. Document changes - Update relevant docs

License

MIT (same as NanoClaw)

Acknowledgments

Based on NanoClaw by qwibitai.