Open Source Plugin-Native Apache 2.0

Agent Platform
for AI-Native Companies

Run autonomous agents that use tools, talk to users, and work together. Self-hosted. Plugin-native. Apache 2.0.

View on GitHub Get Started
Research Agent – Live

Chat, inspect, and iterate after deploy

Two Components, One Product

Start with the Runtime. Add the Hive when you need fleet control.

Required

GolemCore Bot

The Agent Runtime

Run channel-connected agents with plugins, skills, MCP servers, multi-LLM routing, memory, traces, and long-running workflows. Dashboard included.

github.com/alexk-dev/golemcore-bot
Optional

GolemCore Hive

The Orchestrator

Fleet registry, board flows, command dispatch, approval gates, lifecycle signals, and inspection. Coordinates multiple runtimes from a single control plane.

github.com/alexk-dev/golemcore-hive

Quick Start

Get running in under 5 minutes

Install & Start
# One-command install
curl -fsSL https://golemcore.me/install.sh | bash

# Start
cd ~/.golemcore && docker compose up -d

# Open Web UI — configure API keys there
# http://localhost:8080/dashboard
Configure & Extend
# Check logs for temporary admin password
docker logs golemcore-bot

# Open http://localhost:8080/dashboard
# Sign in, add LLM provider keys in Settings
Optional Next Steps
# - Enable Telegram for channel-based chat
# - Install plugins (browser, search, voice, RAG)
# - Connect MCP-backed skills and tool servers
# - Join Hive for control-plane coordination

Why GolemCore

What makes it different from other agent frameworks

Plugin-Native Runtime

Plugins contribute tools, channels, voice providers, and RAG ingestion — they live inside the agent runtime, not outside it.

MCP-Ready by Design

Skills can launch MCP servers. Built-in MCP catalog, dynamic skill factory, and discovery tool for stdio-based tool servers.

Hive Orchestrator

Optional control plane for fleet management. Board flows, command dispatch, approval gates, and inspection across multiple runtimes.

Inspectable Execution

OTel-compatible session tracing, waterfall views, and exported snapshots make agent behavior easy to debug and operate.

Built for Long-Running Work

Delayed follow-ups, progressive memory orchestration, and recovery paths help agents continue past a single turn.

Smart Model Routing

4 model tiers (balanced / smart / coding / deep) with dynamic escalation. OpenAI, Anthropic, Google, and custom endpoints.

Integrations

Services and platforms that work with GolemCore

LLM Providers
OpenAI

GPT, Codex, and reasoning models via the OpenAI API. Default provider with full tool-calling and reasoning support.
Anthropic

Claude Sonnet and Opus models. Extended context windows and strong reasoning for complex multi-step tasks.
Google AI

Gemini models with large context windows. Native multimodal capabilities and competitive pricing.
Channels
Telegram

Built-in Telegram Bot API integration. Chat with your agent from Telegram with user allowlists, webhook bridge, and markdown rendering.
Webhooks
API

Webhook-driven flows for triggering agent sessions from external systems and CI/CD pipelines.
Capabilities
ElevenLabs
TTS

Realistic AI voice generation and text-to-speech. Give your GolemCore bot a natural-sounding voice for audio responses.
Whisper STT
STT

Whisper-compatible speech-to-text. Works with Parakeet, faster-whisper, Whisper.cpp, and any OpenAI-compatible endpoint.
LightRAG
Memory

Knowledge graph-based RAG for long-term memory. Automatic indexing and semantic retrieval across all conversations.
Email (IMAP/SMTP)
Tools

Read, search, and send emails. IMAP and SMTP support with SSL/STARTTLS, threading headers, and HTML sanitization.

From Zero to Running Agent

One install script, one docker compose up, one dashboard

Terminal

Configure API keys, model routing, and plugins from the Web UI

Full Documentation

Ready to Try It?

GolemCore is Apache 2.0 licensed. Clone it, run it, make it yours.

GitHub Docs Issues Discussions