Skill-based architecture with MCP support, tool calling, RAG, and Telegram integration.
Built on Spring Boot and LangChain4j.
// 1. Pull
docker pull ghcr.io/alexk-dev/golemcore-bot:latest
// 2. Run
docker run -d \
-e OPENAI_API_KEY=sk-... \
-e TELEGRAM_ENABLED=true \
-e TELEGRAM_BOT_TOKEN=... \
ghcr.io/alexk-dev/golemcore-bot:latest
Everything you need to build intelligent AI agents in Java
2-stage semantic search + LLM classifier (~5ms + 200ms). Automatic intent matching with fragmented input detection.
Model Context Protocol support for stdio-based tool servers. Connect GitHub, Slack, Google Drive and more out of the box.
Filesystem, Shell, Web Search, Browser, Weather, Skill Management, Goals, Transitions, and DateTime — all sandboxed.
OpenAI, Anthropic (Claude), Google (Gemini), and custom OpenAI-compatible endpoints. Dynamic model tier selection.
LightRAG for long-term memory via knowledge graphs. Automatic indexing and semantic retrieval across conversations.
Autonomous goal-driven execution. The agent plans tasks, executes them, writes diary entries, and sends milestone updates.
Messages flow through 11 ordered systems with up to 20 tool-call iterations
Get running in under 5 minutes
# Configure and run
export OPENAI_API_KEY=sk-proj-...
docker run -d \
--name golemcore-bot \
-e OPENAI_API_KEY \
-v golemcore-bot-data:/app/workspace \
-p 8080:8080 \
--restart unless-stopped \
ghcr.io/alexk-dev/golemcore-bot:latest
# Build
./mvnw clean package -DskipTests
# Configure
export OPENAI_API_KEY=sk-proj-...
# Run
java -jar target/golemcore-bot-0.1.0-SNAPSHOT.jar
# Add Telegram env vars
export TELEGRAM_ENABLED=true
export TELEGRAM_BOT_TOKEN=123456:ABC-DEF...
export TELEGRAM_ALLOWED_USERS=123456789
GolemCore Bot is open source under Apache 2.0. Contributions, issues, and stars are welcome.