# Loadcore

> Loadcore is a distributed control plane that spins up and manages open LLMs across your own GPU servers. One agent per server, one web console for your team, and an OpenAI-compatible API on every server. Prompts and completions never leave your infrastructure.

## Components

### Loadcore Console
- Web app at https://console.loadcore.io
- Manage servers, models, instances, members, departments, invites and API keys
- Search HuggingFace and download GGUF models to a chosen server
- Start and stop llama.cpp instances with port, context size and GPU layers
- In-browser chat with any model the user has access to

### Loadcore API
- Managed multi-tenant control plane (SaaS)
- Organizations, admin and member roles, departments with per-model access
- Server registry fed by agent heartbeats, model and instance state
- Validates personal API keys (prefixed sk-lc-) for agents
- Never proxies inference and never stores prompts or completions

### Loadcore Agent
- Single Go binary per GPU server, no runtime dependencies
- Installs llama.cpp automatically on first start
- Heartbeat every 30 seconds with OS, architecture, GPU, RAM and public IP
- Downloads models to ~/.loadcore/models and reports progress
- Health checks running instances every 15 seconds
- Serves GET /v1/models, POST /v1/chat/completions and POST /v1/completions on port 9090
- Caches API key validation for 5 minutes

## Data path

- Inference: your tools call the agent on your server directly
- Control: the agent sends heartbeats, download progress, instance status and key checks to the API
- Model weights download from HuggingFace straight to your server

## Integrations

Any OpenAI-compatible client: opencode, aider, Continue, Claude Code, LangChain, the OpenAI SDKs, curl.

## Links

- Website: https://loadcore.io
- Console: https://console.loadcore.io
- Documentation: https://loadcore.io/docs
- Model guides: https://loadcore.io/models