Agents
An ok.lol agent is an autonomous AI assistant that runs continuously.
Anatomy of an agent
Every agent has three core resources:
| Resource | Description |
|---|---|
A dedicated address (e.g. agent@ok.lol) for inbound and outbound communication | |
| Wallet | A funded account the agent uses to pay for inference, tools, and external services |
| Tunnel | An encrypted connection between the ok.lol network and your machine |
How agents run
Agents execute on your hardware. ok.lol provides the coordination layer:
- Inbound messages (email, API calls) are routed through the tunnel to your machine
- Your machine runs the agent's logic, calling LLMs and tools as needed
- Outbound actions (sending emails, making payments) go back through the tunnel
- The wallet is debited automatically for any paid resources
Because agents run on your machines, you have full control over the compute environment, installed tools, and data access.
Behavior
Agents are configured with:
- System prompt — defines personality, goals, and constraints
- Tools — functions the agent can invoke (file I/O, web requests, code execution, etc.)
- Memory — persistent context that survives across conversations
All of these can be changed at any time through the dashboard or API.
Proactive agents
Unlike chat-only assistants, ok.lol agents can act on their own:
- Send emails on a schedule or in response to events
- Monitor systems and alert you when something goes wrong
- Write and deploy their own code to extend their capabilities