Cud is a local-first, multi-agent framework designed to be lightweight and intuitive.
Bring the power of autonomous AI agents directly to your machine with Ollama and DeepAgents.
No convoluted setups or bloated abstractions. Intuitive and easy to hack on.
Powered by Ollama. Your data, prompts, and memory stay 100% on your machine.
Create a fleet of specialized agents, each with its own persona and memory.
Behavior is defined in plain Markdown (AGENT.md). No black boxes.
Persistent shell sessions, surgical filesystem operations, and MCP support.
Seamlessly run your agents as background services using systemd.
Extend capabilities with portable skills and MCP servers.
Run scheduled tasks autonomously using cron expressions.
Automatic summarization prevents context limits. Full history is saved to the workspace.
Interact with your agents through Discord slash commands.
Slash Commands
| /new | Start a fresh session context |
| /model <name> | Switch agent's configured model |
| /memory view | View the agent's long-term MEMORY.md |
| /memory clear | Clear the agent's long-term MEMORY.md |
| /usage | Show runtime usage and agent details |
| /reload | Reload tools and system prompt |
Requirements
Ensure you have Ollama running with a tool-calling capable model.
ollama run gemma4:e4b
Installation
pipx install git+https://github.com/arrase/cud.git
Create & Run Agent
Initialize an agent and start the Discord gateway.
cud agent create researcher
cud agent config researcher --model gemma4:e4b
cud gateway setup researcher discord --token YOUR_TOKEN
cud gateway start researcher